[XiaoBinGan]

Problem and result

The Minimal serving runtime job prints the same ResourceWarning lines on every run (#4654). Six open() calls sat
inside an expression, so no close() ever ran:

• serve/telemetry.py — the physical_package_id glob (one handle per CPU), /proc/cpuinfo, /proc/stat,
  /proc/meminfo
• serve/server.py — the two properties files the HIP probe reads

Each now runs under a with statement (or through a one-line _read_lines() helper for the two properties reads).
No value, threshold or return path changes: the same figures are parsed from the same lines in the same order.

Related discussion

#4654 — the report with the exact warning lines and the two sites in server.py. A pytest run with
-W error::ResourceWarning on the CI job is what the report suggests; this PR removes the leaks it names and does not
change the job.

Approach

Only the ownership of the handles changes:

with open(f, encoding="utf-8") as fh:
    packages.add(fh.read().strip())
...
with open("/proc/cpuinfo", encoding="utf-8", errors="replace") as fh:
    for line in fh:
        if line.startswith("model name"):
            return line.split(":", 1)[1].strip()

and for the HIP probe:

def _read_lines(path: str):
    """The lines of `path`, with the file closed when the caller is done (the HIP properties probe reads two)."""
    with open(path) as fh:
        return fh.readlines()

The early return inside the cpuinfo loop is kept; the with closes the file on the way out, which is the bug the
report describes (_cpu_name() reports at the return, not at the open()).

Validation

Passed: python -m pytest serve/test_monitor.py -q on this checkout (Python 3.11.16, Windows): 10 passed, 8
subtests passed — the module these helpers live in imports and its monitor tests are unchanged.

Passed: import serve.telemetry and import serve.server both succeed after the change, and a grep for an open(
still inside an expression in serve/telemetry.py now finds none.

Failed: none.

Not run: the /proc and /sys branches themselves cannot be exercised on this Windows checkout (they are Linux
paths), so the warnings the job reports are verified by reading the diff, not by reproducing them here. A Linux run with
-W error::ResourceWarning would settle it, and the issue's own job is the natural place.

Limits

• No behaviour change is intended, which also means no performance claim: this is about closing handles.
• The two properties reads are now one small helper rather than two identical comprehensions; if the reviewer prefers
  pathlib.Path(...).read_text(...) inline, that is a one-line change each.