feat(rc-admin)!: add bounded client devnull diagnostics

[cxymds]

Related issue

Closes rustfs/backlog#1404

Depends on the bounded diagnostic snapshots merged in #285.

Background and user impact

RustFS beta.10 implements one real active diagnostic performance route: client-to-server upload-to-devnull. Other object, network, and site speedtest families are incomplete or placeholder implementations. rc needs to expose only the measured route with strict load limits, fail-closed capability checks, cancellation, and no automatic retries.

Solution

• Add rc admin diagnostics client-devnull ALIAS --size 8MiB --timeout 30s --concurrency 1 --yes.
• Validate all input before capability discovery or active traffic.
• Limit aggregate traffic to 64 MiB, concurrency to 1-4, and timeout to 1-60 seconds.
• Generate a static zero body incrementally in 64 KiB chunks without allocating the full payload.
• Set exact Content-Length and precompute the matching SHA-256 for SigV4 signing.
• Fail closed on unknown or stubbed capability states before the active POST.
• Never retry an active probe automatically.
• Cancel peer requests and body production on timeout, Ctrl-C, or future abort.
• Bound each active response or error body to 64 KiB.
• Require kind=client-devnull, measured=true, exact rx_bytes, and finite positive duration and throughput.
• Report requested and received bytes, concurrency, wall-clock elapsed time, and aggregate throughput through the existing output-v3 admin_operations family.

BREAKING contract note

BREAKING: This PR updates the protected docs/reference/rc/admin.md behavior contract for a new additive active diagnostics command. It does not change output v1/v2/v3 schemas, config schema, exit-code definitions, or existing command behavior. No schema migration is required.

Validation

Exact head: 73cc41fca3d34edd7bb743e7c871d531a153c673

• cargo fmt --all --check
• CARGO_BUILD_JOBS=2 cargo clippy --workspace -- -D warnings
• CARGO_BUILD_JOBS=2 cargo test --workspace
• Core client-devnull limits: 3 passed
• S3 client-devnull: 9 passed
• S3 read-only diagnostics regression: 3 passed
• Combined CLI diagnostics: 13 passed in both lib and main targets
• CLI request-capture integration: 1 passed
• Cancellation regression: 20 consecutive targeted runs passed
• Review audit: 0 reviews and 0 review threads

Limitations

• Client-to-server devnull only.
• Object speedtest, inter-node network speedtest, site speedtest, site-replication netperf, active drive benchmark, and continuous traffic generation are excluded.
• The RustFS handler itself does not enforce a request-body size limit; this CLI guarantees that it sends no more than 64 MiB, but other callers must enforce their own limits.
• Live configured-environment behavior remains covered by the repository compatibility workflow and GitHub checks; local validation uses source-backed contract and mock integration coverage.

[cxymds]

Validation is complete for exact head e16208271bf7a93b33d1b9b3bb854f2b2b1859ed.

Local gates on the final stacked commit:

• cargo fmt --all --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace

Focused combined validation:

• Core diagnostic contract: 6 passed.
• CLI combined diagnostics: 13 passed in both lib and main targets.
• S3 active probe: 8 passed, explicit abort: 1 passed, read-only diagnostics: 3 passed.
• CLI request-capture integration: 1 passed.
• Independent review, post-fix review, manual-restack review, and final wording review: no remaining findings.

Remote gates:

• CI: https://github.com/rustfs/cli/actions/runs/29852417172 — green, including Linux/macOS/Windows build and tests, MSRV, docs, advisories, Clippy, and protected-file policy.
• Integration Tests: https://github.com/rustfs/cli/actions/runs/29852417350 — green.

The draft PR is open and its true stacked base is restored to cxymds/issue-1400-diagnostic-snapshots (#285). The protected command-reference addition is declared as BREAKING.

A local live beta.10 container run was not performed because the Docker daemon is unavailable in this workspace; the validation claim is limited to the source-backed protocol audit, bounded mock/request-capture tests, local suites, and GitHub checks above.

[cxymds]

CI follow-up for head 73cc41fca3d34edd7bb743e7c871d531a153c673:

The previous macOS test failure was a test synchronization timeout in aborting_client_devnull_closes_an_incomplete_upload. Under runner load, the 5-second channel wait expired before the active request reached the mock server; dropping the test task then caused the secondary "client closed connection before headers" panic. Ubuntu and Windows passed, and no production-path failure was observed.

The fix changes only the two test channel waits from 5 to 15 seconds. Production timeout, cancellation, upload, and traffic-limit behavior is unchanged.

Validation after the fix:

• targeted cancellation test: 20 consecutive passes
• cargo fmt --all --check
• CARGO_BUILD_JOBS=2 cargo clippy --workspace -- -D warnings
• CARGO_BUILD_JOBS=2 cargo test --workspace

The new GitHub checks are now authoritative.