Artifacts.cached never settles waiters when the effect fails, hanging plan #2981

What happens

Artifacts.cached(id) never settles its Deferred when the wrapped effect fails. The first caller gets the failure; every later caller with the same id awaits Deferred.await(deferred) forever.

// packages/alchemy/src/Artifacts.ts
yield* artifacts.set(id, Deferred.await(deferred));
const result = yield* eff;          // on failure, deferred is never completed
yield* Deferred.succeed(deferred, result);

Repro (beta.85):

const once = Effect.fail("boom").pipe(Artifacts.cached("k"));
const first = yield* Effect.exit(once);                                    // Failure
const second = yield* Effect.exit(once).pipe(Effect.timeoutOption("2 seconds")); // None: hung

How this hangs a deploy

During plan, a resource with persisted state is diffed twice: once in resolveResource (stable attributes for downstream consumers) and once in diffResource. Drizzle.Schema caches detectDrift with Artifacts.cached, so when the first diff fails (for example #2980), the second waits forever, and so does every resource downstream of the schema. alchemy deploy sits at "Computing plan" until CI kills it, with no error. In --no-input mode nothing shows which resource is stuck, since failed diffs never emit plan.resource.completed.

Fix we are running

Settle waiters on any exit:

 yield* artifacts.set(id, Deferred.await(deferred));
-const result = yield* eff;
-yield* Deferred.succeed(deferred, result);
+// Settle waiters on failure too, or every later caller hangs.
+const result = yield* eff.pipe(Effect.onExit((exit) => Deferred.done(deferred, exit)));
 yield* artifacts.set(id, result);

(In the TypeScript source the Deferred would need the error type, Deferred.make<A, Err>().) It may also be worth not caching failures at all, so a later caller retries instead of replaying the error.