refactor(test): document and apply conventions

[deciduously]

This PR solidifies testing conventions, applies them to the existing tests, and adds some new functionality to the runner:

• Split multi-behavior tests into individual files per behavior.
• Use consistent style for intent comments, using plain declarative sentences.
• Add intent comments to all nushell and rust unit tests
• Use long argument names.
• Add packages/cli/tests/README.md documenting these conventions.
• Consolidate snapshot normalization into new redact helper, no more PROCESS/<process> confusion.
• Add wait_until helper to replace plain sleep where appropriate.
• Add skip_test helper which reports in a separate category in the summary, instead of early-returning.
• Add skip_if_offline and --offline helper to filter tests which require network.
• Add --release to test against a release build of tangram.
• Add --stress/--stress-count to run tests repeatedly, concurrently with itself, to catch races.
• Add provenance comments citing relevant commits to tests which record a regression/bugfix.
• Refactor runner: use options record instead of positional args, deduplicate pool refill, extract is_failed/round_entries helpers.

No new tests are implemented in this rollup.