fix(ci): enable coverage with thresholds for integration tests

[margaretjgu]

The integration tests were using --disable-coverage which prevented coverage collection but caused tap to exit 1 anyway - tap's built-in report teardown always runs after tap run and exits 1 when .tap/coverage is empty, regardless of --disable-coverage.

This replaces the bypass with actual coverage:

• Removes --disable-coverage
• Adds --coverage-exclude to exclude test files from measurement (we care about library code coverage, not test file coverage)
• Adds --allow-incomplete-coverage since integration tests won't load every source file
• Sets starting thresholds: lines=50, branches=30, functions=60, statements=50

The thresholds are conservative starting points to be tightened once we see actual numbers from the first green CI run. The unit test command (npm test) tracks the same library code at 96/90/96/91 via c8; these are lower because integration tests cover happy paths but not error handling or retry logic.

The OOM from a previous attempt came from wrapping tap with c8. This approach uses tap's native V8 coverage (each test file spawns its own process) which avoids that issue.

[JoshMock]

This LGTM btw! Been sitting in draft for 2 weeks so I kept overlooking it.