fix(webapp): set completedAt for FAILED synthetic span runs

Symmetric fix to the just-applied `ApiRetrieveRunPresenter` change.
PG-resident SYSTEM_FAILURE rows always have `completedAt` populated;
the synthetic span run was returning null for FAILED, so the
run-detail panel and any caller using `isFinished && completedAt` saw
a finished run with no completion timestamp. Fall back to
`run.createdAt` when `isFailed` is true (the buffer entry has no
separate failedAt — createdAt is the best proxy for when the terminal
state landed).

Regression locked in `mollifierSyntheticSpanRun.test.ts` —
the FAILED test now asserts `completedAt === NOW`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: d-cs

apps/webapp/app/v3/mollifier/syntheticSpanRun.server.ts

@@ -102,7 +102,14 @@ export async function buildSyntheticSpanRun(args: {
     updatedAt: run.cancelledAt ?? run.createdAt,
     delayUntil: run.delayUntil ?? null,
     expiredAt: null,
-    completedAt: run.cancelledAt ?? null,
+    // Symmetric with `ApiRetrieveRunPresenter` — FAILED buffered runs
+    // must surface a non-null `completedAt` so the run-detail panel
+    // (and any caller checking `isFinished && completedAt`) doesn't
+    // render a finished run with no completion timestamp. PG-resident
+    // SYSTEM_FAILURE rows always have completedAt set; the buffer
+    // entry has no separate failedAt, so we fall back to createdAt
+    // as the best proxy for when the terminal state landed.
+    completedAt: run.cancelledAt ?? (isFailed ? run.createdAt : null),
     logsDeletedAt: null,
     ttl: run.ttl ?? null,
     taskIdentifier: run.taskIdentifier ?? "",

apps/webapp/test/mollifierSyntheticSpanRun.test.ts

@@ -182,6 +182,12 @@ describe("buildSyntheticSpanRun", () => {
       type: "STRING_ERROR",
       raw: "GATE_REJECTED: buffer rejected the run",
     });
+    // PG-resident SYSTEM_FAILURE rows always have completedAt set;
+    // mirror that on the synth path so callers checking
+    // `isFinished && completedAt` don't render a finished run with
+    // no completion timestamp. The buffer entry has no separate
+    // failedAt, so createdAt is the best-available proxy.
+    expect(synth.completedAt).toEqual(NOW);
   });
 
   it("flags the synthetic run as 'not cached' since cache lookup did not match it", async () => {