fix(aws/scheduler): harden ScheduleGroup reconcile + add lifecycle convergence tests

[sam-goodwin]

Following the SQS hardening template from #184 and the sibling Schedule PR #200. Hardens the EventBridge Scheduler ScheduleGroup resource. Unlike Schedule, schedule groups DO support tagging — so ownership detection, adoption gating, and tag-drift convergence are all in scope here.

Reconciler changes

- if (!observed?.Arn) {
-   yield* scheduler.createScheduleGroup({ Name, Tags })
-     .pipe(Effect.catchTag("ConflictException", () => …listTagsForResource…hasTags…throw…))
-   observed = yield* scheduler.getScheduleGroup({ Name })
- }
- // tag sync diffed against observedTagsResp — kept
+ const arn = observed?.Arn
+   ?? yield* scheduler.createScheduleGroup({ Name, Tags })
+        .pipe(conflictRetry, Effect.map(r => r.ScheduleGroupArn));

• read now returns Unowned(attrs) when the group lacks the alchemy internal tag set, so first-touch adoption requires --adopt / adopt(true) instead of being silently taken over. Subsequent reconciles trust the persisted output.
• Drop the bespoke ConflictException ownership check on createScheduleGroup. The previous code fetched tags, called hasTags, and threw a plain Error if they didn't match — duplicating ownership logic that now lives entirely in read + the engine's --adopt gate.
• Replace it with a bounded ConflictException retry (15 × 2s) on createScheduleGroup, tagResource, untagResource, and deleteScheduleGroup. EventBridge Scheduler returns ConflictException while a group is in the transient DELETING/CREATING state, so re-creating a name through a residual DELETING window now converges instead of failing.
• Re-read the final State from the cloud after reconcile rather than echoing observed?.State (which may be stale by one round-trip).

ConflictException is intentionally NOT tagged RetryableError in distilled — it's context-dependent (DELETING race vs. genuine name collision). The reconciler-level bounded retry stays scoped.

New lifecycle tests

packages/alchemy/test/AWS/Scheduler/ScheduleGroup.test.ts — each test runs destroy → deploy → … → destroy on a ScratchStack and asserts convergence at every step:

• create + delete with default props
• redeploy with same props is a no-op
• reconcile resets tags mutated out-of-band via the raw Scheduler SDK
• reconcile re-creates a group deleted out-of-band (rides the DELETING window)
• changing name triggers replace; old group is deleted
• destroying an already-deleted group is a no-op
• foreign-tagged group requires adopt(true) to take over and rewrites internal alchemy tags

Distilled patch

No patch needed — packages/aws/patches/scheduler.json (per the existing process) already covers ConflictException, and tagging it RetryableError would over-broaden coverage. The reconciler-level bounded retry handles the genuinely transient cases.

[alchemy-version-bot]

Install the packages built from this commit:

alchemy

bun add alchemy@https://pkg.ing/alchemy/2398c87

@alchemy.run/better-auth

bun add @alchemy.run/better-auth@https://pkg.ing/@alchemy.run/better-auth/2398c87

@alchemy.run/pr-package

bun add @alchemy.run/pr-package@https://pkg.ing/@alchemy.run/pr-package/2398c87

[alchemy-version-bot]

Website Preview Deployed

URL: https://alchemyeffectwebsite-worker-pr-230-asjdr4zten6qn7ha.testing-2b2.workers.dev

Built from commit 2398c87.

---

This comment updates automatically with each push.