…are disabled.

…ith stream server and other middleware disabled.

…with stream server re-enabled but still no other middleware.

…a) being actually used and (b) with Box<Dyn>-less post-processing via new `PostprocessingMap` type based on stream::Map, with stream server re-enabled but still no other middleware.

…so fixes bugs in the Cookie middleware relating to adding the COOKIE option more than once, and not preserving OPT header values when appending OPT options.

…er, none of it is needed in the new middleware-as-layered-services world.

…and a stream preceeded by its length doesn't already need its length prepended to it.

- Return Future of Stream from Service::call() instead of the custom Transaction type.
- Relax trait bounds/move trait bounds nearer to the point of use.
- Introduce PostprocessingStream impl of Stream for stream based upstream result post-processing.
- Introduce ServiceResult aka io::Result.
- Modify the serve_zone example to assume it is powered by (not included in these changes) XFR middleware.

…ng the domain server functionality.

…567_agent`".

…ce'".

…t the identity case.

…al I/O errors occur (such as the other end has disconnected) and for not aborting response stream processing if there is no space in the write queue. These issues were hit when testing with the default dig timeout of 5 seconds and a large zone that took a while to start transferring.

…very many Tokio tasks, otherwise we compete with other actions being performed by the applicaition. Instead use a dedicated thread pool for zone walking for AXFR transfers. Also use a Tokio unbounded receiver stream instead of a ordered futures stream to avoid creating N boxed futures, to avoid a mutex lock around the stream, and to avoid waiting for the initial zone walk to finish before sending any AXFR records.

…ool crate.

…imited support for A/IXFR in/out and NOTIFY, e.g. doesn't honor RETRY or EXPIRE SOA timers yet, and needs a lot of cleaning up. Has been seen successfully exchanging zones with NSD via XFR acting both as primary and secondary.

Relevant changes:
- Adds net::client::auth: a new client layer for doing TSIG request signing and response validation.
- Extends Catalog to use net::client::auth.
- Add Message::is_streaming() and ComposeRequest::is_streaming() which returns true if the QTYPE of the first question indicates a type of request which may cause a stream of responses instead of a single response, limited to AXFR and IXFR at the moment.
- Adds GetResponse::stream_complete() and GetResponse::is_stream_complete() to signal from a caller that the last message of a response stream has been seen (as only the caller can know this, the DNS protocol does not include a general purpose end marker for response streams if the TCP stream is kept open for further use) as TSIG stream validation needs to know when the last message of the stream has been encountered.
- Factors the code of ComposeRequest::to_message() out to the call sites (cache.rs, dgram.rs) and removes ComposeRequest::to_vec() as they were not compatible with net::client::auth.
- Extends Request::is_answer() to NOT require AXFR subsequent responses to have a question (as it is optional per RFC 5936).
- Extends net::client::stream support more than a single response to a single request:
  - Adds Config::set_streaming_response_timeout() and Config::set_initial_idle_timeout().
  - Adds Request::stream_complete.
  - Extends send_request() to behave differently if ComposeRequest::is_streaming() is true.
  - Preserves
- Extends the serve-zone example to pass a key store with a hard-coded HMAC-SHA256 TSIG key to the Catalog.

Unrelated changes:
- Also extend the serve-zone example to persist a zone to disk on commit, and optionally load it at startup.
- Removes an unnecessary problematic Clone bound on dgram.impl SendRequest for Connection in dgram.rs.
- Adds impl Display for net::client::stream::ConnState.
- Logs at trace level the reason a net::client::stream is closed.
- Logs failed handling of NOTIFY requests and returns an error to the client instead of panicking.
- Logs XFR-in progress.
- Replace the large match data block in Catalog with much simpler code.

… summary.

…ded by Stelline XFR tests (to make requests using the Stelline client instead of a real client).

…n. Currently seems to break some other Stelline tests.

- Batch as many RRs per AXFR response as will fit.
- Support backward compatible AXFR out (one RRset per response) if configured for the client IP.
- Reject AXFR over UDP unless we're (a) transferring the entire zone for IXFR because there is no diff available, and (b) (TODO) the response fits in a single datagram.
- Disable the TCP idle timer while in a transaction to avoid timing out XFR responses (especially in Tokio paused time test mode but could also occur normally).
- Honour stream shutdown in the Stelline ClientServerChannel to avoid a never ending test.
- Use 127.0.0.1 as the client in Stelline server tests, not ::.
- Support EXTRA_PACKET in Stelline SECTION ANSWER blocks for streaming response matching.
- Add an initial IXFR Stelline test.

- Replace thread pool based concurrency limits with semaphores instead.
- Extract batching functionality to owner RrBatcher type and also use it for IXFR.

- Move client specific XFR settings for compatibility mode into the existing client specific XFR settings stored by the Catalog.
- Enforce XFR ACLs for XFR out.
- FIX: Don't panic if a Stelline CHEKCK_ANSWER block doesn't specify an answer section.
- FIX: Allow more than one provide-xfr config setting in a Stelline server config block.
- Extend the Stelline parser to take an optional COMPATIBLE keyword for the provide-xfr server config setting.
- Extend the auth_xfr.rpl Stelline test to test AXFR compatibility mode.

Known issue: The new compatible AXFR test fails due to an issue with the "QUERY ADDRESS <IP>" not being used. This is fixed already in PR #336..

- Factor XFR client code out.
- Pass XFR settings around instead of looking them up in multiple places.
- More access control enforcement.
- More RustDoc comments.
- Remove outdated/incorrect comments.
- Add required missing Sync bound.

- Replace blanket Service impl for fn with  blanket Service impl for Deref to avoid requiring a service to be an Arc.
- Auth client no longer needs to take an Arc<Key>.
- WIP TSIG signing on xfr-in responses.
- Remove the RwLock used by the Catalog on the key store as it doesn't edit the store.

… playing with.

  Enables response signing to be the last thing that happens as required by the RFC.
- Introduce RequestMeta and PostProcessingMeta, used to pass the detected TSIG key name from TsigMiddlewareSvc to XfrMiddlewareSvc.
- Temporarily disable Clippy lint that requires RustDocs to be fully written.
- Add enable(bool) fn to CookiesMiddlewareSvc and EdnsMiddlewareSvc to make them usable in passthrough only mode, to support conditional service chain construction.
- Rename Svc to NextSvc to make it clearer which service is referred to.
- Minor error handling & code readability improvements in XfrMiddlewareSvc.
- Fix Xfr with TSIG - verified with NSD, Bind 9 and Dig.
- Known issue: If TSIG doesn't fit on a TCP streaming response, the code will panic.

…eptional case.

…ing batching of RRs into responses during AXFR processing that we don't forget to leave space for the TSIG RR or ENDS OPT RR.

- Fixed a bug in TSIG client sequence verification introduced by an earlier fix for server sequence signing (also pushed to PR 345).
- Fixed issues with AXFR response batching.
- Fixed a Stelline test bug where a request for a UDP client was treated as a request for a TCP client.
- Fixed an errant blank line per zone entry bug in processing of the `local-data` Stelline server config setting.
- Fixed a zone walking bug that omitted CNAMEs at non-leaf nodes (also pushed as PR #352).
- Made the zone argument to net::client::xfr optional.
- Made net::client::xfr pass on the messages it receives after it has checked/processed them.
- Added a multi-response large zone variant of the AXFR test.
- Added a Stelline TSIG variant of the AXFR test.
- Renamed the XFR Stelline tests.
- Renamed the TSIG client from net::client::auth to net::client::tsig.
- Added the NOTIFY middleware back in to the server Stelline tests.
- Added the TSIG middleware to the server Stelline tests.
- Extended the server Stelline tests to:
  - Create a 'TESTKEY' in the key store.
  - Wrap TCP connections created by the client factory in the tsig and xfr clients.
  - Wrap UDP connections created by the client factory in the tsig client.
  - Accept a named TSIG key in the `provide-xfr` server config setting.
  - Accept a named TSIG key in the `STEP N QUERY` Stelline command.
  - Take the entire ENTRY as input to the client factory, not just the client IP address.
    (so that any specified TSIG key for the client can be configured)
  - Accept `$ORIGIN` in `ANSWER SECTION` response RRs.
    (so that many RRs with a shared origin can be expressed more succinctly)
  - Change in-memory zone tree node ordering to be deterministic (BTreeMap based instead of HashMap based) in #[cfg(test)] mode.
    (so that the new large AXFR response test can know which response RRs will be present in which response message)
- Removed unused stream connection ID numbering and related Tokio tracing span that was causing unwanted test output indentation.
- Removed some unnecessary calls to the zone walking callback op fn.

…g the NOTIFY opcode in upcoming tests).

…ently.

Add command line arguments to the serve-zone example.
TSIG sign and verify NOTIFY requests too.

Changed and fixed the data structures and logic for XFR diffs, also enabling received IXFRs to be kept as diffs.
Changed how a connection factory is passed to the Catalog to enable it to be replaced with Stelline transports when testing.
Renamed ZoneType to ZoneConfig.
Add support for "allow-notify" and "request-xfr" in Stelline server configs.
Add support for specifying the Stelline request OPCODE.
Added a get_rrset) fn to WritableZoneNode.