fix(actions): make validator sync atomic to stop dropped validators

[lucca65]

Problem

upsertAction replaced an action's validators by deleting on db (a separate connection, unawaited) and inserting on tx (also unawaited). The transaction committed before the inserts completed, and the out-of-transaction delete could race and wipe the just-inserted rows. Result: actions left with zero validators. A claimable action with no validators is invisible to validators in the app, so its claims sit pending forever.

A manual reindex on 2026-06-13/14 replayed cambiatus.cm::upsertaction as updates (action_id > 0) and tripped this race at scale.

Impact (prod, read-only audit 2026-06-29)

• 262 claimable actions reindexed 06-13/14 with 0 validators
• 187 stuck pending claims, ~80 users, across 8 communities (AGL + MUDA worst hit)
• Surfaced via a user report (validators couldn't see the claims to approve them)

Fix

Run the delete + inserts inside tx, awaited via Promise.all, with the whole chain returned so the transaction commits only after they land. Drop the per-insert swallowing .catch so any failure rolls the action back instead of persisting a half-written action. Re-running the indexer over the affected blocks now repopulates validators reliably (also makes future reindexes safe).

Follow-up (separate, not in this PR)

Backfill the already-broken 262 actions: re-process the affected upsertaction blocks with this fix deployed, then verify 0 actions with missing validators and the 187 claims become visible.

Testing

• npx standard clean.
• No automated suite in this repo. Behavior validated by static analysis of the tx/promise ordering; recommend a local indexer replay of an upsertaction update against a dev DB before deploy.

🤖 Generated with Claude Code