Per-test output capture

Associate captured stdout/stderr with individual tests instead of interleaving everything.

Currently Python `print()` output from parallel tests gets mixed together. Capturing per-test and displaying it alongside failures (like pytest's default capture behavior) would make output much more readable.