fix(soak): close Phase 6 / Phase 7 24-h gates + pin daemon log strings

[githubrobbi]

Why

The 2026-05-11 Phase 6 + Phase 7 24-h soak runs (LOG/uffs_soak/phase6-20260509-213122 and LOG/uffs_soak/phase7-20260510-214412) each landed 1 failure in the validator — and in both cases the daemon was actually doing the right thing. The soak harness's log-scrape just couldn't see it.

Phase 6 — Drive C warm_ttl_sec exceeds peers failed at 300s vs 300s

The post-EMA-fix daemon was computing warm_ttl_sec ≈ 3687s for drive C every tick during the synthetic-load window — the Phase 6 adaptive-bonus formula engaged correctly. The validator's parse_max_ttl_field just couldn't see it because crate::index::transitions::evaluate_idle_demote only emits a DEBUG-level shard.ttl event when it actually proposes a demote. During the synthetic load drive C sits in Warm/Hot with idle_secs ≈ 0, so the demote-eval ladder never reaches the DEBUG-level idle-demote / min-tier-clamp arms. The catch-all (None, _) arm fires at TRACE — and the harness's RUST_LOG=shard.ttl=debug filtered it out.

Fix (scripts/dev/long-soak.rs:746): shard.ttl=debug → shard.ttl=trace. ~23 k extra trace events over 24 h (~3.5 MB), marginal against the existing 75 MB log volume.

Phase 7 — Encrypted-cache refresh fired during soak failed at found 0 refresh-class events

The validator hunted for USN refresh tick|trigger_save|threshold.*save|encrypted cache refresh — none of those alternatives match the daemon's actual INFO line:

```
INFO Journal poll: triggered background compact-cache save drive=F reason=AgeElapsed cursor=151008
```

Fix (scripts/dev/long-soak.rs:1244): regex re-anchored on compact-cache save.

Retroactive verification: the existing 24-h Phase 7 log contains 11 such events:
```
$ grep -c 'compact-cache save' LOG/uffs_soak/phase7-20260510-214412/daemon.log
11
```

Daemon-side regression pins

Two new tests so a future log-message rename fails CI before reaching a 24-h soak gate:

• crate::cache::journal_loop::tests::save_log_message::compact_cache_save_log_message_pins_string_target_and_level — pins target = uffs_daemon::cache::journal_loop, level = INFO, and the literal compact-cache save substring on the journal-loop save event.
• crate::index::tests::shard_ttl_events::below_ttl_event_pins_target_level_message_and_reason — pins target = shard.ttl, level = TRACE, message = Adaptive idle-demote evaluation: not yet idle past TTL, reason = below-ttl, and the four TTL fields the harness's parse_max_ttl_field depends on.

Both tests reuse the existing crate::index::tests::tracing_capture EventLog scaffold. Visibility hoisted from pub(super) → pub(crate) on the scaffold items + pub(crate) on mod tests; in index/mod.rs and mod tracing_capture; in index/tests/mod.rs. No production-code visibility changes — only sharing of an already-test-only scaffold across crate-internal modules.

Verification

```
cargo test --package uffs-daemon --lib # 283 + 2 = 285 pass
cargo clippy --package uffs-daemon --tests # clean
cargo fmt --package uffs-daemon -- --check # clean
just lint-fast # green (ran via pre-commit hook)
just lint-pre-push # green (ran via pre-push hook)
```

What's NOT in this PR

• No production-code behavior change. The daemon's log messages, levels, and event fields are unchanged — only newly pinned by tests.
• No new dependencies.
• No CI workflow / just-recipe changes.

Operator next steps after merge

1. Pull main on the Windows soak host.
2. --dev smoke (5 min each):
   ```
   rust-script scripts/dev/long-soak.rs phase6 --dev
   rust-script scripts/dev/long-soak.rs phase7 --dev
   ```
3. Re-run the 24-h gates:
   ```
   just soak phase6
   just soak phase7
   ```

Closes the Phase 6 / Phase 7 follow-up gates documented in docs/architecture/memory-tiering-windows-host-validation.md.