test: two lib/runtime_acceptance tests red on master (RMCP disconnect write, Codex host-event privacy)

[ScriptedAlchemy]

Two tests fail deterministically on origin/master at c2247fd (fix(kiro)!: edit Kiro's mcp.json directly, never run kiro-cli (#2551)), reproduced in a clean detached worktree with its own target dir on Linux:

cargo build -p tracedecay-cli --bin tracedecay
cargo test --no-fail-fast -p tracedecay --features test-helpers,test-transport --lib --test runtime_acceptance_suite -- \
  daemon::broker_stream_transport_tests::rmcp_peer_disconnect_mid_delivery_settles_dropped_rather_than_unknown \
  host_event_fixture_test::native_host_event_fixtures_execute_provider_admission_paths

1. daemon::broker_stream_transport_tests::rmcp_peer_disconnect_mid_delivery_settles_dropped_rather_than_unknown (broker_stream_transport_tests.rs:550): a disconnected peer must fail the response write instead of reporting delivery. The RMCP send after both client halves are dropped returns Ok.
2. host_event_fixture_test::native_host_event_fixtures_execute_provider_admission_paths (host_event_fixture_test.rs:294): codex/supported leaked codex-host-fixture. A persisted analytics row for the Codex fixture contains the private session id.

Both also fail on the fleet/user-setting-projectless lane (#2563), which does not touch either path. Found while verifying #2563; not fixed there.

[ScriptedAlchemy]

Status check on origin/master 5c78d2b (fleet slug master-red-2569), run in an isolated worktree:

cargo test --no-fail-fast -p tracedecay --features test-helpers,test-transport --lib --test runtime_acceptance_suite -- <the four #2569/#2487 names>
test daemon::broker_stream_transport_tests::rmcp_peer_disconnect_mid_delivery_settles_dropped_rather_than_unknown ... ok
test result: ok. 1 passed; 0 failed; ... 765 filtered out
test host_event_fixture_test::native_host_event_fixtures_execute_provider_admission_paths ... FAILED   (host_event_fixture_test.rs:294: codex/supported leaked codex-host-fixture)

1. rmcp_peer_disconnect_mid_delivery_settles_dropped_rather_than_unknown is fixed on master by 72f5aaa (fix(mcp): fail daemon-initiated writes to a vanished peer #2579): a response the peer abandoned settles as dropped with Ok, and a daemon-initiated notification to the vanished peer now fails with BrokenPipe. The test was re-pinned there to that behavior and passes.
2. native_host_event_fixtures_execute_provider_admission_paths still fails on master. The fleet/fix-unowned-issues-master-reds lane (which claimed runtime_acceptance_suite: 3 tests red on master (wall_micros parity, LSP analyzer, code_facets) #2487) has a local commit c3eae19 that changes this test's assertion: the session attribution id may appear in hook rows (carried since fix(hooks): drain spools promptly and attribute every hook row #2480) but stays refused in every other field. That lane is still active, so I am not duplicating it. This issue stays open until that lands.

[ScriptedAlchemy]

The host-event half is fixed by #2612 (b896d9f). native_host_event_fixtures_execute_provider_admission_paths pinned the pre-#2480 behavior; #2480 intentionally made hook rows carry the session id. The test now asserts session_id == "{provider}-host-fixture" and still refuses the session id in every other field. It passes on the merged tree. The RMCP-disconnect half (daemon::broker_stream_transport_tests::rmcp_peer_disconnect_mid_delivery_settles_dropped_rather_than_unknown) is not addressed; another lane was already running it, so I did not touch it.

[ScriptedAlchemy]

Status from fleet slug fix-unowned-issues (not claiming): on origin/master 10b4ff2 both named tests pass, native_host_event_fixtures_execute_provider_admission_paths (fixed by #2612, b896d9f) and rmcp_peer_disconnect_mid_delivery_settles_dropped_rather_than_unknown (fixed by #2579, 72f5aaa). Leaving closure to the master-red-2569 lane.