test(mcp): e2e integration coverage for #63892 poll-loop OOM spin

The salvaged unit tests hand-construct completed futures to lock the
_run_on_mcp_loop contract. Add a live-loop test that reproduces the actual
field trigger: an inner asyncio.wait_for expiry stores a real TimeoutError
on a real future scheduled on the MCP loop. Asserts the fixed loop surfaces
it once, promptly -- not spinning to the outer deadline (which both leaked
memory and masked the real error behind the generic wrapper message).

tests/tools/test_mcp_poll_loop_oom_integration.py

@@ -0,0 +1,68 @@
+"""End-to-end coverage for the MCP poll-loop OOM spin (#63892).
+
+The unit tests in ``test_mcp_tool.py`` hand-construct completed futures to lock
+the ``_run_on_mcp_loop`` contract deterministically. This complements them by
+exercising the actual field trigger through the *live* MCP loop: an inner
+``asyncio.wait_for`` expiry produces a real ``TimeoutError`` stored on a real
+future scheduled via ``run_coroutine_threadsafe``. On Python >= 3.8 that class
+is the builtin ``TimeoutError``, so before the fix the poll loop swallowed the
+completed future's stored exception and spun with no sleep -- burning CPU,
+growing the exception traceback ~108 MB/s until the gateway OOM'd, and finally
+masking the real error behind the generic "MCP call timed out after <full
+timeout>" wrapper.
+
+The fixed loop must surface the real exception once, promptly -- long before
+the outer deadline.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import time
+
+import pytest
+
+
+@pytest.fixture
+def mcp_loop():
+    import tools.mcp_tool as mcp_tool
+
+    mcp_tool._ensure_mcp_loop()
+    yield mcp_tool
+    mcp_tool._stop_mcp_loop()
+
+
+def test_inner_wait_for_timeout_surfaces_promptly_without_spinning(mcp_loop):
+    async def inner():
+        # Real TimeoutError from wait_for, completing far before the outer
+        # deadline below -- the exact shape of an MCP call_tool wrapped in the
+        # server's configured mcp_servers.<srv>.timeout.
+        await asyncio.wait_for(asyncio.sleep(60), timeout=0.05)
+
+    start = time.monotonic()
+    with pytest.raises(TimeoutError) as exc:
+        mcp_loop._run_on_mcp_loop(inner, timeout=10)
+    elapsed = time.monotonic() - start
+
+    # Fixed: the real inner TimeoutError surfaces within a couple poll ticks.
+    # Broken: the loop swallows it and spins with no sleep until the 10s
+    # deadline, then raises the generic wrapper message instead. A generous
+    # (>= 2s) bound keeps this stable on a slow CI runner while still failing
+    # hard on a spin-to-deadline regression.
+    assert elapsed < 5.0, (
+        f"poll loop spun for {elapsed:.1f}s instead of resolving the completed "
+        "future once (#63892 regression)"
+    )
+    assert "MCP call timed out after" not in str(exc.value), (
+        "the real inner TimeoutError must surface, not the outer poll deadline"
+    )
+
+
+def test_successful_call_still_returns_through_real_loop(mcp_loop):
+    """Guard the done-future path against regressing normal success returns."""
+
+    async def inner():
+        await asyncio.sleep(0.25)  # first polls time out while still pending
+        return {"ok": True}
+
+    assert mcp_loop._run_on_mcp_loop(inner, timeout=10) == {"ok": True}