Measure platform-specific process paths in the coverage gate

Decision: Valid test-measurement gap.

Evidence: `pyproject.toml` excludes both `if sys.platform == "win32"` and `if sys.platform != "win32"` suites from coverage reporting. `zuvloop/_process.py` also excludes substantial libuv process code with pragmas that say macOS CI exercises it. Windows runs tests without coverage.

Impact: A platform path can stop executing while the aggregate report still claims 100% branch coverage.

Next action: Collect coverage artifacts on Linux, macOS, and Windows and combine them in one aggregate gate. Limit exclusions to genuinely unreachable defensive branches with explicit reasons.

Acceptance criteria: Every supported platform-specific implementation line is measured on at least one platform, and the combined branch report retains `fail_under = 100`.

Start by inspecting the platform exclusions in pyproject.toml and the coverage pragmas in zuvloop/_process.py, then review the existing Linux, macOS, and Windows test jobs. Configure coverage artifacts to be combined into one report, retaining only justified exclusions; done means supported platform-specific lines are measured and the aggregate branch gate keeps fail_under = 100.

Written by the indexing model from the issue text.