[quad341]

What this changes

cairn's internal index-freshness check (ensureFresh) is split into two
variants. Six read paths (Find, FindCorrection, Status, Search,
Recall, CullCandidates) now use a best-effort variant: if the bounded
self-heal reindex times out, they serve the (possibly lagging) index instead
of failing the call. The one path that gates a mutation (EntryByID, used by
cull-evict, promote-mark, and anchor-attach) is unchanged and still requires
a fully fresh index before proceeding.

Separately, Find's "not found" fallback — which force-reindexes to confirm
a real miss — was previously unbounded and could stall for the full
lock-contention window (~33s). It's now bounded by the same timeout the
self-heal path already uses, and reports "not found" (rather than a timeout
error) if that bound is hit — a real miss and an unconfirmable-within-budget
miss now look the same to callers.

Review notes

• internal/cairn/evict.go's EntryByID intentionally still calls the
  strict variant — it's the sole pre-mutation gate for delete, tier
  promotion, and anchor-attach, and should not be relaxed.
• The new bound on Find's not-found fallback reuses the existing
  self-heal timeout constant rather than introducing a new one.
• No new logging, no dependency changes, no config surface added.

Test plan

• go build ./... && go test ./... -race -count=1 — all 7 packages
  green, zero failures/skips.
• 3 new tests cover: a best-effort caller serving a lagging index on
  reindex overrun, EntryByID still hitting the strict deadline
  (regression guard), and a genuinely-missing id returning "not found"
  within the bound instead of stalling.
• Release gate: release-gates/crn-tk8qd-gate.md

🤖 Deployed by actual-factory