Observe storage errors before journal and timer decisions

[PLN]

Storage-error formatting held shared journal or timer decision locks. A delayed retirement error could block capture/diagnostic admission and retention metadata, while timer load/save errors could block disarm, rearm and status. Both stores now observe error messages before taking their admission/decision mutex, then validate current ownership and publish the same diagnostics.

The journal regression fails on the prior implementation and now preserves independent capture, diagnostics, retained-error reporting and cleanup retry. Both timer load/save regressions also fail before the fix; delayed errors no longer block independent arm decisions or publish into a different arm. Twenty combined journal/timer observation repetitions passed.

Final validation: full uncached race suite passed (manager 59.973s, journal 42.594s, timers 3.305s), plus vet, Windows/Linux staticcheck, complete diff/privacy review and local documentation links. Exact source c4f6341b55176940358736b462d624db0acc1828 passed CI 34768748133 and the consolidated 57-case matrix: 50 manager, four journal and three timer cases, three repetitions per SYSTEM/headless standard-user identity (8.680s/6.613s). No cases skipped; process, linger and profile cleanup passed. All three binary hashes and source identity were verified in retained evidence. Full R2 acceptance remains open, including notification-reader admission.

The initial candidate's broader lab run stopped when the PowerShell wrapper treated expected native stderr as a terminating script error (three completed cases passed, no test failures observed). Its partial evidence is retained; it is not qualification. The corrected wrapper retains stderr and checks every test-process exit code explicitly. A local three-run stderr rehearsal passed before the fresh final-source fixture was staged.