mason: anchor marker coverage mutation on its fence assertion

The coverage-guard mutant now fails at expect(fences).toBe(0): bypassing pending.ordinal > consumed.ordinal reaches beforeApply and arms the fence for a newer pending target. Baseline and mutant both leave pendingReads at 2; the preceding read-count assertion is only source context in Bun's failure frame, not the failing assertion.

Keep the exact named test and use only its exact fence assertion as expect_message; do not widen the match or change production code or assertions. Linux Bun 1.4.2 baseline passes, and a controlled guard mutation fails only that named test with Expected: 0 / Received: 1 at the fence assertion; the source was restored byte-for-byte.

The workflow-pinned ck-mutate 0.7.0 (commons 7d08e73722fa3e79bbcc2607753978ab768f1c6b) was installed, but local catalogue check/replay cannot acquire its Git-metadata lock on the read-only Linux snapshot. Per parent instruction, do not create shadow Git metadata; leave authoritative CAUGHT confirmation and the full catalogue check to the Mutations workflow.

Co-authored-by: Alfonso <alfonso-magic-context@users.noreply.github.com>

Author: ualtinok

mutations.toml

@@ -201,7 +201,7 @@ runner = 'command'
 command = ['bun', 'test', 'packages/plugin/src/hooks/magic-context/transform-postprocess-phase.test.ts', '--timeout', '30000', '-t', '{test}']
 test_count_pattern = 'Ran {count} test'
 expect_red = ['^deferred\scompaction\smarker\srepresentation\snewer\spending\spublication\swaits\suntil\sconsumed\sby\sthe\sserved\sresponse$']
-expect_message = 'expect(getPersistedCompactionMarkerState(db, sessionId)).toBeNull()'
+expect_message = 'expect(fences).toBe(0)'
 
 [[control]]
 id = 'ts-marker-cas-clear-keeps-newer-pending'