fix: use Mapping instead of dict for EvalResult.scores

[starfolkai]

Summary

EvalResult.scores is typed as dict[str, float | None], but dict is invariant in its value type, so callers constructing an EvalResult with a homogeneous dict[str, float] (no None values) get a type error from both pyright and mypy:

error: Argument of type "dict[str, float]" cannot be assigned to parameter "scores"
  of type "dict[str, float | None]" in function "__init__"
  Type parameter "_VT@dict" is invariant, but "float" is not the same as "float | None"
  Consider switching from "dict" to "Mapping" which is covariant in the value type

This PR widens the annotation from dict[str, float | None] to Mapping[str, float | None]. The framework only iterates result.scores.items() in build_local_summary and never mutates scores after construction, so Mapping is the correct abstraction. Same fix pattern as #281 (EvalParameters) and #285 (Metadata / ParametersSchema). Mapping is already imported.

Fixes #292.

Test plan

• pyright on the issue's repro snippet: 0 errors, 0 warnings
• mypy on the issue's repro snippet: Success: no issues found
• pyright + mypy on src/braintrust/type_tests/: clean
• pytest src/braintrust/test_framework.py: 35 passed, 1 skipped
• ruff check + pre-commit hooks on framework.py: clean
• End-to-end Eval(...) against the live Braintrust API with mixed float / None scorer outputs: results and summary unchanged ({'exact_match': 1.0, 'always_none': None}, summary 100%)

🤖 Generated with Claude Code