test: combine operational pressure and storage faults

[PLN]

Combines native path-trigger bursts, repeated process failures, large stdout/stderr, stalled journal readers and storage faults in one isolated manager workload. The ordinary case injects a partial write followed by disk-full errors; the opt-in case fills a validated small disposable volume. Both assert bounded resources, responsive status/stop, retained cleanup, record recovery and no trigger resurrection over three cycles.

Validation:

• Exact source 3605c2c7978c4b10076558a76e5453fd051e60bc passed Windows/Linux CI 34813927032, including race suites, vet, staticcheck, vulnerability checks and builds. Final local journal race tests passed in 64.469 seconds; repeated aggregate/combined pressure tests and the earlier full local race suite passed.
• Windows 11 Enterprise LTSC 26100.9168 qualification passed 19 selected tests three times per SYSTEM/headless standard-user identity, without skips. Each identity ran nine injected-fault and nine actual disk-full combined cycles, plus three standalone exhaustion/recovery cases. Suites took 55.701 and 53.236 seconds. One logical CPU was enforced; the combined subprocess uses GOMAXPROCS=2.
• Worst measured status / concurrent stop / post-repair cleanup: SYSTEM 11.691 / 1053.758 / 205.930 ms; standard user 27.419 / 1064.887 / 254.639 ms. Every cycle retained the 16 MiB / 16,384-record queue budget and reported drops. All native memory/worker/handle budgets passed; final manager subprocesses retained three goroutines.
• Fixture processes, profiles and linger grants were released; the disposable VHD detached and the original hosting broker remained running. Exact artifact hashes, logs and failed-attempt evidence are retained privately.

The volume fixture truncates and flushes its owned filler before deletion, then verifies returned capacity. Earlier native runs demonstrated that deleting the filler could leave available space at zero for over 30 seconds, invalidating the assumed start of recovery. Runtime deadlines remain unchanged.

Only test fixtures change. R5.5 acceptance and measured bounds will be recorded in a following documentation change; overall R5 still needs diagnostics/event resources.