fix: bound the async admission call so a wedged worker cannot hang callers

[puzza007]

The admission handshake from the exactly-once dispatch work (#178) used an infinity gen_server:call, so a wedged worker — typically its C port stopped draining the pipe and port_command blocked the worker mid-callback — silently absorbed async_req callers with no bound at all. This was the one regression surface of making dispatch a synchronous call.

Admission is a port write plus a map insert, microseconds when healthy, so the call is now bounded at 5s: a worker that cannot answer in time yields an immediate {error, #{code => admission_timeout}} and the caller can retry — re-rolling onto a healthy random worker — instead of blocking indefinitely. A request the worker admits after the caller gave up is cleaned up by its own request timer. The fail-fast contract is uniform: a timeout during overload-spillover also returns immediately, so an admission waits at most one bound, never one per candidate.

The bound applies only to the async admission ack, derived from the message kind in one place (admission_bound/1): a synchronous request's single call carries the whole response — the worker replies when the transfer or its request timer completes — so it stays unbounded at the call layer and is governed by the request timeout as before. (The first cut bounded both paths and the suite caught it immediately: 120 sync tests truncated at 5s.) attempt/2 guards its timeout mapping on the bound, since on an unbounded call a {timeout, _} exit can only be a worker that died with that exit reason — without the guard, sync callers could see the async-only error code.

Deterministic coverage in katipo_async_failure_SUITE: a suspended worker (the messages-queue-but-nothing-runs shape of the production wedge) must produce admission_timeout within the bound. 199 tests, dialyzer, xref, and lint pass.

Top of the stack: #180 → #182 → #183 → #184 → this. Merge bottom-up; I'll rebase each layer as the one below lands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Vf2gxZLpxT4HvBfZKCgP3g

[codecov]

Codecov Report

❌ Patch coverage is 92.30769% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 78.23%. Comparing base (046fce5) to head (bec37ad).

Files with missing lines	Patch %	Lines
src/katipo.erl	92.30%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #185      +/-   ##
==========================================
- Coverage   78.31%   78.23%   -0.09%     
==========================================
  Files           9        9              
  Lines        1365     1369       +4     
==========================================
+ Hits         1069     1071       +2     
- Misses        296      298       +2     

Flag	Coverage Δ
c	78.23% <92.30%> (-0.09%)	⬇️
erlang	78.23% <92.30%> (-0.09%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.