feat(test-set): add configurable sampling methods

[jakepresent]

Summary

• Ports the Omni #395 sampling redesign into ASSERT test-set generation.
• Adds sampling under pipeline.test_set.prompt / scenario with stratified, full_factorial, and random methods.
• Defaults to stratified-by-behavior sampling for parity with Omni, while keeping tiny smoke runs valid through stratum subsampling.
• Builds this on top of the current science/viewer PR stack so it can be discussed as the remaining science-parity delta.

Notes

• This changes test-set generation semantics, so I am opening it as draft for scope review rather than treating it as an automatic release merge.
• Depends on the current science-sync/viewer stack: science: sync structured systematization output #111 and fix(viewer): show scenario run model labels #120.

Validation

• python -m pytest tests/test_sampling.py tests/test_test_set.py tests/test_test_set_stage.py tests/test_systematization_stage.py tests/test_systematization_convert_stage.py tests/test_viewer_metrics.py tests/test_viewer_run_page_server.py -q
• python -m pytest tests/test_cli.py tests/test_schema_contracts.py -q
• git diff --check