TIKA-4815: trivial follow-ups from the PR #3022 review

[tballison]

What's in it:

• PipesClientInterruptTest — setStartupTimeoutMillis(1000) bounds the non-interruptible handshake read that made the backoff test's flake window 60 s; both
  tests now try-with-resources the client; @timeout 30→45 (above the 35 s sum of awaits).
• TikaGrpcConcurrencyTest — both burst tests now assert pipesParser.isSharedMode() matches the config, so a useSharedServer binding regression is catchable.
• PerClientServerManager — log no longer says "mid-request" (it also fires mid-handshake) or claims an immediate recycle.
• grpc index.adoc — Concurrency section ported from the README.

[tballison]

@dpol1 wdyt?

[dpol1]

the isSharedMode pin is the right call - the burst alone can't prove the mode - and bounding the handshake read kills the one real flake window. LGTM