feat: aggregate expect.soft() failures per test

[Skn0tt]

Summary

• Moves soft assertion collection from fixture teardown to test call phase using pytest_runtest_call hook.
• At end of test: 0 failures → noop, 1 failure → re-raise as FAILED, ≥2 failures → raise ExceptionGroup as FAILED.
• Eliminates spurious "passed" entries when soft assertions fail—the test is correctly reported as FAILED, not passed+failed.
• Requires playwright-python with the soft-assertions hook (feat(assertions): add expect.soft() for collecting multiple failures playwright-python#3065).
• Backward-compatible with older Playwright versions that lack soft-assertions support.

Companion PR (core hook): microsoft/playwright-python#3065

Example output

When running tests with one single soft failure, multiple soft failures, one passing case, and one soft+body failure case:

• Single soft failure is reported as FAILED with the assertion details (e.g. expected 'goodbye', actual 'hello').
• Multiple soft failures are reported as FAILED with an ExceptionGroup containing each soft failure.
• Matching soft assertions pass cleanly.
• A body exception plus a soft assertion failure reports both in a grouped ExceptionGroup as FAILED.

FAILED tests/test_soft.py::test_soft_plus_body_failure - ExceptionGroup: Test and soft assertion failures (2 sub-exceptions)
FAILED tests/test_soft.py::test_single_soft_failure - AssertionError: Locator expected to have text 'goodbye'
FAILED tests/test_soft.py::test_multiple_soft_failures - ExceptionGroup: Soft assertion failures (2 sub-exceptions)
PASSED tests/test_soft.py::test_soft_passes

[dgozman]

I wonder what happens if the test has a soft failure followed by the hard failure? Are we going to see both errors?

[Skn0tt]

yes, see test_soft_assertion_does_not_shadow_body_failure

[dgozman]

The code looks good. It would be nice to include pytest output in the description, to see how it looks like for the user.