[AdaAibaby]

Relationship to #3606 (please read before merging). #3608 and #3606 are an intentionally disjoint, complementary split of the Redis-storage observability series — not duplicates. They can merge in either order.

• obs(api): add Redis storage layer metrics for lock, transition, scan, and index #3608 (this PR) owns the lock / transition / scan metrics plus api.redis_storage.team_index.size. It deliberately does not declare expiration_index.size or expiration_index.sweep_duration.
• obs(api): add expiration index size and evictor sweep duration metrics #3606 owns exactly those two expiration-index metrics. Nothing is declared in both PRs, so they do not collide at merge.

Summary

Implements the 8 observability gaps identified in #3607, plus the index-size and sweep-duration instrumentation from the related issues (#3605 / #3606).

Lock (lock.go)

• api.redis_storage.lock.wait_duration ms histogram — total wait in Obtain per successful acquisition, via a new obtainLock wrapper on Storage. Replaces the two direct s.locker.Obtain call sites in state_change.go.
• api.redis_storage.lock.timeout_total counter — Obtain calls whose context expired before the lock was held.

Transition (state_change.go)

• api.redis_storage.transition.duration ms histogram — wall-clock from StartRemoving setting the transition key until the callback writes the result (transitionStart captured before the lock, passed into the callback closure).
• api.redis_storage.transition.execution_mismatch_total counter — both ErrExecutionMismatch paths in StartRemoving (early refusal + Lua script refused).
• api.redis_storage.transition.callback_failures_total counter — callback closures that returned early because the callback lock was not obtained (waiters receive no result notification on this path).
• api.redis_storage.transition.joined_total counter — StartRemoving calls that joined an in-flight transition to the same target state.
• api.redis_storage.transition.wait_fallback_polls_total counter — each 1-second ticker iteration in waitForTransition (PubSub notification missed or dropped).

Scan (scan.go)

• api.redis_storage.scan.teams_skipped_total counter — per-team failures silenced by forEachSandboxBatch's continue.
• api.redis_storage.scan.corrupt_records_total counter — records failing json.Unmarshal in fetchSandboxBatch.

Index size / sweep (heal.go, items.go)

• api.redis_storage.team_index.size {entry} histogram — per-team SCARD pipelined in recordTeamIndexSizes, once per heal pass (5 min). p99/max detects unbounded growth.
• api.redis_storage.expiration_index.size {entry} histogram — global expiration ZSET ZCARD, piggybacked on the same heal pass.
• api.redis_storage.expiration_index.sweep_duration ms histogram — wall-clock of each ExpiredItems (evictor scan) call.

All metric names registered in packages/shared/pkg/telemetry/meters.go with descriptions and units.

Closes #3607

Test plan

• go build ./packages/api/internal/sandbox/storage/redis/... passes (verified locally)
• go vet ./packages/api/internal/sandbox/storage/redis/... ./packages/shared/pkg/telemetry/... clean (verified locally)
• Integration tests pass in CI (requires Docker / Redis testcontainer)
• After deploy: api.redis_storage.lock.wait_duration, transition.duration, expiration_index.sweep_duration histograms visible in Grafana
• Confirm transition.execution_mismatch_total stays near-zero under normal traffic; spikes on execution-pinned kill races