test(live): retain load context for timing failures (iter 264)

[ractive]

The three timing tests now report measured wall/internal timing and host load after the timed operation, so a failure retains useful saturation context. Their existing 2500 ms, 2000 ms and ±750 ms bounds remain unchanged; load never waives a failure.

Iteration 264 records ten isolated and ten controlled CPU-load samples per test (60 actual measurements), per-test dispositions, and a failing dishonest-elapsed mutation followed by byte-for-byte product restoration. Controlled CPU load does not reproduce every aspect of the historical concurrent-browser workload. All three original acceptance criteria are fulfilled.

Validation

• Independent local review: explicit zero findings; one review, no repair rounds.
• Current stable 1.98.1; cargo fmt → strict workspace/all-target clippy → workspace tests passed on unchanged final test source.
• All nine enumerated xtask gates passed; absent dogfood script/Firefox references are recorded skips, not live evidence.
• FF_RDP_LIVE_TESTS=1 FF_RDP_LIVE_NETWORK_TESTS=1 cargo run -p xtask -- live-sweep: 346 unique named tests passed across all six tiers (335+1+3+3+2+2).

LIVE_SWEEP_SUMMARY executed=346 skipped=0 preexisting=0 vanished=0 launch_timeout=0 timed_out=0 total=346
LIVE_SWEEP_PROFILES leaked=0 unattributed=0 root=/Users/james/Library/Application Support/ff-rdp/profiles

All owned load workers and the raw sweep browser/profile were cleaned up; desktop Firefox was preserved. Evidence: .git/ralph-loop/20260919-queue/iter264/validation-record.md, raw per-command logs/metadata, supervisor-accounting.json, and iter264-review/phase-result.json. Actual token usage is unavailable.

Carry-over

None produced by this iteration. Historical timing observations remain in the plan; parked iteration 203 and unrelated daemon failures are not claimed fixed.