fix: self-heal in-flight counts by pruning stale request ids

On a long-lived SSE connection a missed remove leaves a stale request id in
the tracker forever; llama-swap only sends a fresh snapshot on connect, so
counts drifted up and stayed stuck. The tracker now timestamps each request
id (upserts refresh it) and prunes ids not updated within 120s, on every
inflight event and on a 30s runtime interval. Removes also accept numeric
ids defensively.

tests/event-feed.test.ts

@@ -67,6 +67,15 @@ test("decodeEvent parses an inflight remove with a bare id", () => {
   assert.deepEqual(ev, { type: "inflight", operation: "remove", id: "817" });
 });
 
+test("decodeEvent accepts a numeric id on remove", () => {
+  const msg: SseMessage = {
+    event: "message",
+    data: JSON.stringify({ type: "inflight", data: JSON.stringify({ operation: "remove", id: 820 }) }),
+  };
+  const ev = decodeEvent(msg);
+  assert.deepEqual(ev, { type: "inflight", operation: "remove", id: "820" });
+});
+
 test("decodeEvent parses modelStatus", () => {
   const msg: SseMessage = {
     event: "message",

tests/inflight-tracker.test.ts

@@ -73,3 +73,34 @@ test("total sums in-flight requests across all models", () => {
   tracker.apply({ type: "inflight", operation: "remove", id: "1" });
   assert.equal(tracker.total(), 2);
 });
+
+test("prune drops request ids that stopped receiving updates", () => {
+  let now = 0;
+  const tracker = new InflightTracker(() => now);
+  tracker.apply({ type: "inflight", operation: "add", requests: [{ model: "A", id: "1" }] });
+  now = 200_000;
+  tracker.apply({ type: "inflight", operation: "add", requests: [{ model: "A", id: "2" }] });
+  tracker.prune(60_000);
+  assert.equal(tracker.total(), 1);
+  assert.equal(tracker.count("A"), 1);
+});
+
+test("active requests refreshed by upserts survive pruning", () => {
+  let now = 0;
+  const tracker = new InflightTracker(() => now);
+  tracker.apply({ type: "inflight", operation: "add", requests: [{ model: "A", id: "1" }] });
+  now = 200_000;
+  tracker.apply({ type: "inflight", operation: "add", requests: [{ model: "A", id: "1" }] });
+  tracker.prune(60_000);
+  assert.equal(tracker.total(), 1);
+});
+
+test("prune is applied on every inflight event", () => {
+  let now = 0;
+  const tracker = new InflightTracker(() => now);
+  tracker.apply({ type: "inflight", operation: "add", requests: [{ model: "A", id: "stale" }] });
+  now = 200_000;
+  tracker.apply({ type: "inflight", operation: "add", requests: [{ model: "B", id: "live" }] });
+  assert.equal(tracker.total(), 1);
+  assert.equal(tracker.count("B"), 1);
+});