feat(seasonpass): bump mainnet image to git-887fe57 (deploy #308 GQL-out-of-txn + #306 pool 60)

[ipdae]

What

Bumps the mainnet seasonpass.image.tag from git-ba5228c to git-887fe57.

#308 (fix/gql-outside-db-txn) is stacked on #306, so this single tag bump ships both on top of the currently-live ba5228c:

commit	change	PR
485b172	raise DB pool 30 → 60 (absorb connection holds)	#306
887fe57	call cleared-stage RPC outside the DB transaction	#308

(Live ba5228c already had cc9e3f8 async /ping + ba5228c anyio threadpool cap.)

Why

Repeated block-status / invalid-claim alarms (Assertible → PagerDuty / #9c-mainnet-alert). Root cause caught live via pg_stat_activity: a status request held a user_season_pass row lock idle-in-transaction for 150s+ while doing a slow get_last_cleared_stage GQL call. That piled up claim SELECT ... FOR UPDATE waiters (up to ~1300s), drove the SQLAlchemy pool to 30/30, and timed out every DB endpoint → both monitors tripped together.

#308 moves the GQL call out of the transaction in the three status paths (get_default_usp WORLD_CLEAR, user_status/all_user_status exp==0 refetch): capture the needed scalars, rollback() to release the connection/lock, then do the RPC. #306's pool 60 is the headroom backstop.

Review

code-reviewer verdict: merge — no blocking issues on the production money-path.

• ✅ Verified: no DetachedInstanceError on happy path (FastAPI serializes before session teardown; expired-but-attached objects lazy-reload). season_pass= → season_pass_id= safe. No write loss. outer try/except handles RPC timeout without connection leak.
• 🟡 non-blocking follow-ups: (1) narrow rollback()→commit() concurrent-delete window can 500 on ObjectDeletedError (low probability — these rows are rarely deleted); (2) expire_on_commit lazy-reload adds extra checkouts at serialization (net-positive vs the long holds).
• 🟢 minor: loop-var pass_type shadowing; status-endpoint test gap.

Migrations

None coupled. The #307 partial index (idx_claim_unsettled) is already applied live, and this code works with or without it. Migrations are applied manually (no init-container/Job).

Deploy

ArgoCD 9c-external-services has no auto-sync — after merge, manual sync required:

kubectl -n argocd patch applications.argoproj.io 9c-external-services --type merge \
  -p '{"operation":{"initiatedBy":{"username":"admin"},"sync":{"revision":"HEAD"}}}'

🤖 Generated with Claude Code