feat: whisper-side cancellation + SSE-triggered retry

- Add cancelJob() to whisper.ts: sends DELETE /jobs/:id to the whisper
  server (best-effort, errors silently ignored)
- DELETE /api/jobs/[id] now calls cancelJob() when cancelling an active
  job that has a whisperJobId, stopping GPU use immediately
- Webhook handler guards against locally-cancelled jobs: returns ok early
  so whisper's late completion cannot overwrite cancelled status or send
  a phantom 'Transcript ready' notification
- Replace blind sleep(Retry-After + 1s) in submitJob() with
  waitForModelReady(): subscribes to /model/events SSE and proceeds as
  soon as state:ready arrives; falls back to the Retry-After timeout if
  SSE is unreachable or closes without model_ready
- Refactor retry tests to use URL-aware makeJobFetch() helper; add 7 new
  tests (3 SSE-triggered retry, 3 cancelJob, 1 webhook cancelled-guard)
  — 144/144 passing

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

src/lib/server/whisper.ts

@@ -4,8 +4,6 @@ function whisperUrl() {
 	return process.env.WHISPER_URL ?? 'http://localhost:8080';
 }
 
-const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
-
 /** Get the current model state from whisper-rtx2080. */
 export async function getModelStatus(): Promise<ModelStatus> {
 	const { default: fetch } = await import('node-fetch');
@@ -16,11 +14,69 @@ export async function getModelStatus(): Promise<ModelStatus> {
 	return res.json() as Promise<ModelStatus>;
 }
 
+/**
+ * Wait for the whisper model to become ready.
+ *
+ * Subscribes to /model/events SSE and resolves as soon as a payload with
+ * state:"ready" arrives. Falls back to a plain timeout (`timeoutMs`) if the
+ * SSE connection fails or closes without that event, so the retry loop can
+ * try again without hanging indefinitely.
+ */
+async function waitForModelReady(timeoutMs: number): Promise<void> {
+	const { default: fetch } = await import('node-fetch');
+	const ac = new AbortController();
+	return new Promise((resolve) => {
+		let done = false;
+		const finish = () => {
+			if (!done) {
+				done = true;
+				ac.abort();
+				resolve();
+			}
+		};
+
+		const timer = setTimeout(finish, timeoutMs);
+
+		fetch(`${whisperUrl()}/model/events`, { signal: ac.signal as AbortSignal })
+			.then(async (res) => {
+				if (!res.body) {
+					clearTimeout(timer);
+					return finish();
+				}
+				let buf = '';
+				for await (const chunk of res.body) {
+					if (ac.signal.aborted) break;
+					buf += chunk.toString();
+					const lines = buf.split('\n');
+					buf = lines.pop() ?? '';
+					for (const line of lines) {
+						if (!line.startsWith('data:')) continue;
+						try {
+							const payload = JSON.parse(line.slice(5).trim());
+							if (payload.state === 'ready') {
+								clearTimeout(timer);
+								finish();
+								return;
+							}
+						} catch { /* ignore parse errors */ }
+					}
+				}
+				// Stream ended without model_ready → proceed to retry immediately
+				clearTimeout(timer);
+				finish();
+			})
+			.catch(() => {
+				// SSE unreachable — fallback timer will fire eventually
+			});
+	});
+}
+
 /**
  * Submit an audio file to whisper-rtx2080. Returns the whisper job id.
  *
- * Handles 503 (model not ready) transparently: retries using the
- * `Retry-After` header until the model loads or maxAttempts is exhausted.
+ * Handles 503 (model not ready) transparently: retries after subscribing to
+ * /model/events SSE — proceeds as soon as state:"ready" arrives, or after the
+ * Retry-After timeout elapses (whichever comes first).
  * Calls `onModelWaiting` on each 503 so the caller can surface the wait to the user.
  */
 export async function submitJob(
@@ -60,7 +116,7 @@ export async function submitJob(
 			const state = body.state ?? 'unloaded';
 			const waitSecs = body.retry_after_secs ?? parseInt(res.headers.get('Retry-After') ?? '15');
 			onModelWaiting?.(state, waitSecs);
-			await sleep((waitSecs + 1) * 1000);
+			await waitForModelReady((waitSecs + 1) * 1000);
 			continue;
 		}
 
@@ -71,6 +127,20 @@ export async function submitJob(
 	throw new Error(`Whisper model did not become ready after ${maxAttempts} attempts`);
 }
 
+/**
+ * Cancel a queued or running job on the whisper server (best-effort).
+ * Errors are silently ignored — local job status is already set to cancelled.
+ */
+export async function cancelJob(whisperJobId: string): Promise<void> {
+	try {
+		const { default: fetch } = await import('node-fetch');
+		await fetch(`${whisperUrl()}/jobs/${whisperJobId}`, {
+			method: 'DELETE',
+			signal: AbortSignal.timeout(5000)
+		});
+	} catch { /* best-effort */ }
+}
+
 /** Open an SSE stream from whisper and call onProgress/onDone callbacks. */
 export async function streamJob(
 	whisperJobId: string,