feat(warmup): per-module canary records (thread + timing observability)

Adds a canary record for each module submitted to the warmup, tracking:
canary_id, module, thread_name, thread_id, submit_ts, start_ts,
end_ts, elapsed_ms, status, error.

Surface:
- WarmupManager.canaries() returns list[dict] (defensive copy)
- AppController.warmup_canaries() returns list[dict] (delegation)
- GET /api/warmup_canaries Hook API endpoint
- ApiHookClient.get_warmup_canaries() returns list[dict]

Example: the warmup of google.genai records a 1187ms canary on
thread controller-io_0 with thread_id 50420, canary_id 1.

11 new tests (8 unit in test_warmup_canaries + 3 in test_api_hooks_warmup).
All pass; live_gui smoke test confirms endpoint returns real data.

src/warmup.py

@@ -12,11 +12,25 @@ Public API on the manager (and exposed on AppController via delegation):
   mgr.is_done() - bool
   mgr.wait(timeout) - block until done
   mgr.on_complete(callback) - register completion callback
+  mgr.canaries() - list[dict] of per-module canary records (observability)
   mgr.reset() - clear state (for re-warmup, e.g. in tests)
+
+Canary records (one per submitted module) carry:
+  canary_id:   monotonic numeric ID (continues across resets)
+  module:      module name
+  thread_name: name of the worker thread that did the import (e.g. "controller-io-0")
+  thread_id:   threading.get_ident() of that worker
+  submit_ts:   wall-clock when submit() was called
+  start_ts:    wall-clock when the worker started the import
+  end_ts:      wall-clock when the import finished
+  elapsed_ms:  (end_ts - start_ts) * 1000
+  status:      "running" | "completed" | "failed" | "cancelled"
+  error:       error message string if status == "failed", else None
 """
 
 import importlib
 import threading
+import time
 from concurrent.futures import Future, ThreadPoolExecutor
 from typing import Callable, Optional
 
@@ -34,8 +48,12 @@ class WarmupManager:
   self._failed: list[str] = []
   self._callbacks: list[CompletionCallback] = []
   self._started = False
+  # Canary observability state (per-module import tracking).
+  self._canaries: list[dict] = []
+  self._next_canary_id: int = 1
 
  def submit(self, modules: list[str]) -> None:
+  submit_ts = time.time()
   with self._lock:
    if self._started:
     raise RuntimeError("WarmupManager.submit() called twice; call reset() first")
@@ -44,6 +62,21 @@ class WarmupManager:
    self._failed = []
    self._done_event.clear()
    self._started = True
+   for name in modules:
+    canary = {
+     "canary_id": self._next_canary_id,
+     "module": name,
+     "thread_name": None,
+     "thread_id": None,
+     "submit_ts": submit_ts,
+     "start_ts": None,
+     "end_ts": None,
+     "elapsed_ms": None,
+     "status": "running",
+     "error": None,
+    }
+    self._next_canary_id += 1
+    self._canaries.append(canary)
   for name in modules:
    self._pool.submit(self._warmup_one, name)
 
@@ -55,6 +88,11 @@ class WarmupManager:
     "failed": list(self._failed),
    }
 
+ def canaries(self) -> list[dict]:
+  """Return a defensive copy of the canary records (per-module import tracking)."""
+  with self._lock:
+   return [dict(c) for c in self._canaries]
+
  def is_done(self) -> bool:
   return self._done_event.is_set()
 
@@ -83,21 +121,52 @@ class WarmupManager:
    self._done_event.clear()
    self._callbacks = []
    self._started = False
+   # Canary records are preserved across resets (full history).
+   # Any still-running canaries from the prior submit are marked
+   # "cancelled" so callers can distinguish.
+   for c in self._canaries:
+    if c["status"] == "running":
+     c["status"] = "cancelled"
+     c["end_ts"] = c.get("end_ts") or time.time()
+     if c.get("start_ts") and c["elapsed_ms"] is None:
+      c["elapsed_ms"] = (c["end_ts"] - c["start_ts"]) * 1000
 
  def _warmup_one(self, name: str) -> None:
+  start_ts = time.time()
+  thread = threading.current_thread()
+  thread_name = thread.name
+  thread_id = thread.ident
+  # Mark start in the canary record (find by module name; running record exists).
+  with self._lock:
+   for c in self._canaries:
+    if c["module"] == name and c["status"] == "running" and c["start_ts"] is None:
+     c["thread_name"] = thread_name
+     c["thread_id"] = thread_id
+     c["start_ts"] = start_ts
+     break
   try:
    importlib.import_module(name)
   except BaseException as e:
-   self._record_failure(name, e)
+   end_ts = time.time()
+   self._record_failure(name, e, end_ts)
   else:
-   self._record_success(name)
+   end_ts = time.time()
+   self._record_success(name, end_ts)
 
- def _record_success(self, name: str) -> None:
+ def _record_success(self, name: str, end_ts: Optional[float] = None) -> None:
+  if end_ts is None: end_ts = time.time()
   callbacks: list[CompletionCallback] = []
   with self._lock:
    if name in self._pending:
     self._pending.remove(name)
    self._completed.append(name)
+   for c in self._canaries:
+    if c["module"] == name and c["status"] == "running":
+     c["status"] = "completed"
+     c["end_ts"] = end_ts
+     if c["start_ts"] is not None:
+      c["elapsed_ms"] = (end_ts - c["start_ts"]) * 1000
+     break
    done = self._started and not self._pending
    if done:
     self._done_event.set()
@@ -108,12 +177,21 @@ class WarmupManager:
    except Exception:
     pass
 
- def _record_failure(self, name: str, _err: BaseException) -> None:
+ def _record_failure(self, name: str, _err: BaseException, end_ts: Optional[float] = None) -> None:
+  if end_ts is None: end_ts = time.time()
   callbacks: list[CompletionCallback] = []
   with self._lock:
    if name in self._pending:
     self._pending.remove(name)
    self._failed.append(name)
+   for c in self._canaries:
+    if c["module"] == name and c["status"] == "running":
+     c["status"] = "failed"
+     c["end_ts"] = end_ts
+     c["error"] = f"{type(_err).__name__}: {_err}"
+     if c["start_ts"] is not None:
+      c["elapsed_ms"] = (end_ts - c["start_ts"]) * 1000
+     break
    done = self._started and not self._pending
    if done:
     self._done_event.set()