2.2.0

Needs Vega 1.1: Garuda depends on it from: "1.1.0".

• Idle connections give their buffers back. An HTTP/1.1 or HTTP/2
  connection that has been quiet for --idle-buffers milliseconds (100;
  ServerConfig.idleBufferDelay) returns its read buffer to Vega's receive
  blocks and its output buffer to a per-worker list, and waits in Vega's
  receive(), which holds no buffer, until it is sent more. A WebSocket
  reader does so as soon as it has nothing buffered. With 20,000 idle
  keep-alive connections a connection costs 14.5 KiB of the server's memory
  where it cost 27.0, and with 5,000 over TLS 18.3 KiB where it cost 41.2.
  The limits a connection is held to are shared, not copied into each of its
  waiting calls.
• --idle-buffers MS|off. 0 gives the buffers back as soon as a
  connection has nothing to read; off (nil) keeps them, as 2.1 did. The
  100 ms default keeps a busy connection, whose next request is usually a
  fraction of a millisecond away, from giving its buffers back and taking
  them again for every request: doing that cost 3–7% of throughput on the
  keep-alive workloads, and with the delay they measure as 2.1 does.
• The README opens with a handler that awaits a PostgreSQL query and throws
  an HTTPError, which is what most handlers look like, where every snippet
  before the middleware section used to be an async closure that never
  suspended. The quick start is a five-line application with no database.
  The opening is shorter: the runtime, shards and crash semantics are one
  paragraph with links, the middleware list is grouped by purpose, the
  database, streaming and upload prose is cut to what a first reader needs,
  and the axum side-by-side, the release line and the Kitura note move to
  the sections about them.
• The README says that Garuda 2 is rebuilt on Vega, a new I/O and runtime
  package modelled after Tokio, in place of aviancore.