port/audit upstream PR 2284: fix(analytics): preserve distance invalidation during an active build #622

Port or audit of upstream Kpa-clawbot/CoreScope PR 2284 ("fix(analytics): preserve distance invalidation during an active build"). Read the upstream PR and its linked issue for details (read only; never comment there).

What upstream fixed

Upstream: background-fill completion could reset an active sync.Once, causing a fatal mutex unlock or letting an older distance build mark an incomplete index ready. Fix: build admission under the existing mutex plus a generation check.

What to do here

1. Audit first. Find out whether our fork has the same defect. Our code has diverged in several places; do not cherry-pick blindly.
2. If it applies: port the fix adapted to our code, with a red test before and green after, plus a mutant that the test kills. Include perf numbers where the change touches a hot path.
3. If it does not apply: add tests that pin the correct behaviour (as in audit: store lock reads and region/area set resolution vs upstream PRs 2246 and 2247 #597/audit: store lock readers vs upstream 2246/2247 #602) and explain why the defect cannot occur here.

Where to look

We made distance indexing lazy (#558/#589) and chunked. Check whether our lazy build, background fill loader and invalidation can race the same way (stale completion consuming an invalidation, or a Once reset while active). Use -race and a deterministic interleaving test.

Constraints: no SQL under PacketStore.mu; cmd/server stays read-only; no new map[string]interface (Kpa-clawbot#1383); CSS variables only for UI changes.