fix: report output format failures as usage errors

[codeforester]

Summary

• Classify public OutputFormatError failures at the run_app process boundary as exit-code 2 usage errors.
• Preserve safe actionable dependency/format messages in human output and use output_format_error in JSON mode.
• Add process-boundary tests for optional YAML dependency failures and document the contract.

Closes #343

Validation

• uv run --extra dev --extra typer --extra quality python -m pytest tests/test_optional_yaml_dependency.py tests/test_json_contracts.py -q
• Ruff check and format check for changed Python files
• git diff --check

[codeforester]

Consistency: the new except OutputFormatError branch never checks reraise_unexpected, unlike every other exception branch in run_app() that offers it as a debugging/testing escape hatch (e.g. base_cli.testing.invoke(app, args, reraise_unexpected=True)). Callers relying on that pattern to inspect the raw exception/traceback can't do so for this new error path.

[codeforester]

Additional finding (file not touched by this diff, so not postable as an inline comment):

lib/python/base_cli/_lifecycle.py:150 — outcome_from_exception() has no case for OutputFormatError, so the new usage-error classification this PR adds in run_app() isn't mirrored in run history/telemetry/diagnostics. Reproduced: a command raising OutputFormatError (e.g. missing PyYAML) correctly returns exit code 2 to the shell/JSON caller, but the persisted run.json lifecycle bundle records "outcome": "unexpected_error", "exit_code": 1, "status": "error" — a real on-disk divergence from what was actually returned — and _record_lifecycle_diagnostic() logs it as an 'Unexpected command exception' with a full traceback, mislabeling a clean, documented usage error as an internal crash. Any tooling that inspects run history/telemetry (e.g. basectl runs list) will disagree with the actual exit code. Recommend adding an OutputFormatError case to outcome_from_exception() alongside this fix.