Harden timing-sensitive tests: replace Process.sleep with proper synchronization

[ppsplus-bradh]

Summary

Replace Process.sleep calls across 3 test files with deterministic synchronization, eliminating a class of flaky test failures on fast or slow CI runners.

Motivation

Several tests use fixed Process.sleep(50-100) calls to wait for async operations (request delivery, stream cleanup, process shutdown, supervisor restarts). These are brittle — too short on slow runners, unnecessarily slow on fast ones. The codebase already has MockCLI.poll_until/2 for proper async polling; these tests just weren't using it.

Note on timeouts

MockCLI.poll_until/2 has a default timeout of 5 seconds and a polling interval of 10ms. Both can be overridden via options (timeout:, interval:) to shorten the overall polling period where appropriate. If the condition is never met, the test fails with a raised exception rather than hanging indefinitely.

Changes

test/claude_code/session_test.exs

Test	Was	Now
"stream cleanup removes request" (adapter)	Process.sleep(50) x2	Direct :sys.get_state assert (sync call guarantees state) + MockCLI.poll_until for cleanup
"discards raw message for unknown request_id"	Process.sleep(50)	Synchronous :sys.get_state call
"logs warning for unparseable raw message"	Process.sleep(50)	Synchronous :sys.get_state call
"stream cleanup removes request" (streaming)	Process.sleep(100) + Process.sleep(50)	Direct :sys.get_state assert + MockCLI.poll_until for cleanup
"handles multiple concurrent queries"	Process.sleep(100)	MockCLI.poll_until on map_size(state.requests)
"handles sequential streaming queries"	Process.sleep(100)	MockCLI.poll_until on map_size(state.requests)

test/claude_code_test.exs

Test	Was	Now
"returns false for stopped session"	Process.sleep(100)	Process.monitor + assert_receive {:DOWN, ...}
"stops a running session"	Process.sleep(100)	Process.monitor + assert_receive {:DOWN, ...}

test/claude_code/supervisor_test.exs

Test	Was	Now
"restarts an existing session"	Process.sleep(50)	MockCLI.poll_until for new pid
"restarts crashed sessions automatically"	Process.sleep(100)	MockCLI.poll_until for new pid
"handles multiple session crashes independently"	Process.sleep(100)	MockCLI.poll_until for count + pid change

Additionally, two supervisor tests were corrected after validation revealed their premises were no longer accurate:

Test	Was	Now	Reason
"handles session startup errors gracefully"	Process.sleep(100) + assert length(children) <= 1	Renamed to "handles session startup with empty config", asserts count == 1	Empty config [] is valid — api_key is now optional (defaults to ANTHROPIC_API_KEY env var). The child starts successfully and stays alive; it never crashes. The original test assumed a missing api_key would cause a crash, which is no longer the case.
"handles invalid session config"	Process.sleep(50) + assert count >= 0	Renamed to "starts session with empty config", asserts count == 1 and Process.alive?(pid)	Same reason — testing the inverse (child survives) validates current behavior.

Test plan

• All 1509 tests pass
• mix quality passes (compile, format, credo, dialyzer)
• Zero Process.sleep calls remain in the 3 modified files
• Code review verified polling conditions are correct and no no-op polls remain
• CHANGELOG.md updated

[ppsplus-bradh]

CI Note: The test failure (health/1 during provisioning in port_integration_test.exs) is a pre-existing race condition that is already fixed in #42. This PR is dependent on #42 merging first.

Once #42 is merged, this branch will need to be rebased onto the updated main for CI to pass.

If #42 is rejected or closed, the fix for that test can be brought into this PR directly.

[guess]

LGTM - we can merge after getting in #42

[guess]

@ppsplus-bradh can you rebase this please 🙏