fix: reject empty scorer prompt sets

[Bortlesboat]

Summary

Empty scorer prompt sets now fail during scorer initialization with a clear error instead of producing a later scoring crash.

KeywordRate previously accepted 0 prompts and could later divide by zero. KLDivergence could proceed to baseline-logit collection with no prompts. Both built-in scorers now reject empty prompt sets immediately after loading them.

Validation

• PYTHONPATH=src uv run python -m unittest discover -s tests -p test_scorers.py
• uv run ruff format --check src/heretic/scorers/keyword_rate.py src/heretic/scorers/kl_divergence.py tests/test_scorers.py
• uv run ruff check --extend-select I src/heretic/scorers/keyword_rate.py src/heretic/scorers/kl_divergence.py tests/test_scorers.py
• uv run ty check src/heretic/scorers/keyword_rate.py src/heretic/scorers/kl_divergence.py tests/test_scorers.py
• git diff --check

[gemini-code-assist]

Code Review

This pull request adds validation to ensure that both the KeywordRate and KLDivergence scorers are initialized with at least one prompt, raising a ValueError otherwise. It also introduces corresponding unit tests to verify this behavior. I have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[red40maxxer]

This also looks like a reasonable change to me, though I'm not sure if the unit tests are strictly necessary.

[Bortlesboat]

They're cheap now that CI runs the tests directory, but I don't feel strongly — happy to drop them and keep just the validation if you'd rather keep the suite lean.

[Vinay-Umrethe]

This also looks like a reasonable change to me, though I'm not sure if the unit tests are strictly necessary.

yes I don't think that test is necessary, it did not changed anything affecting reproducibility. And tests are already passing for what cases we have as of now.

[Bortlesboat]

Dropped the tests in 219aaee — the PR is now just the two validation checks.

Heads up that CI is red on Python 3.10 only, with gemma-4e failing the
model.safetensors hash check. That's unrelated to this change: the file removed
here was a unittest case using a stub context that never loads a model, and
run_tests.py only globs SHA256SUMS.* per model directory. 3.11, 3.12, and 3.13
all pass on the same commit.