Centralize ONNX export warning suppression in export_pytorch()

[Copilot]

After removing eager TracerWarning filtering at startup, several ONNX export paths began leaking TracerWarning/export-time noise to stderr because suppression existed only in one caller. This change moves suppression to the shared export implementation so all call sites behave consistently.

• Export path hardening

  • Moved warning suppression into src/winml/modelkit/export/pytorch.py::export_pytorch().
  • Wrapped exporter invocation with warnings.catch_warnings() + warnings.filterwarnings("ignore") so all transitive callers (CLI export, HF build path, perf path, etc.) inherit suppression.

• Caller cleanup

  • Removed the now-redundant local warnings wrapper from src/winml/modelkit/commands/build.py.
  • Keeps behavior but eliminates per-caller duplication/footgun.

• Documentation alignment

  • Updated src/winml/modelkit/_warnings.py note to reflect that suppression is now handled in export_pytorch() rather than build.py.

• Focused regression coverage

  • Added tests/unit/test_pytorch_warning_suppression.py to verify:
    • warning context is actually entered in export_pytorch()
    • warnings emitted during export do not propagate to callers

# src/winml/modelkit/export/pytorch.py
with warnings.catch_warnings():
    warnings.filterwarnings("ignore")
    return exporter.export(...)

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• releases.astral.sh
  • Triggering command: /home/REDACTED/.local/bin/uv /home/REDACTED/.local/bin/uv run ruff check (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)