Add LEXam public leaderboard converter

[JoelNiklaus]

Add LEXam public leaderboard converter

Problem

Every Eval Ever supports converting local eval logs and some public leaderboards (e.g. AlpacaEval), but not LEXam - a legal reasoning benchmark with a public leaderboard and Hugging Face dataset.

Solution

Add a lexam converter that:

• Fetches the public leaderboard HTML from the LEXam project website repo
• Parses open-question judge scores and MCQ accuracy for each model
• Emits schema-valid Every Eval Ever JSON with source_metadata.source_type = documentation
• Links underlying benchmark data to LEXam-Benchmark/LEXam on Hugging Face
• Records LLM-judge metadata for open-question scoring
• Writes output under the stable data/lexam/{developer}/{model} layout
• Uses explicit model identity mappings and fails loudly on newly unmapped leaderboard entries

CLI usage:

uv run every_eval_ever convert lexam --output_dir data

Testing

• Added unit tests with a frozen HTML fixture covering parsing, medal stripping, metric combination, schema validation, missing-section errors, and unknown model mappings
• Ran live conversion against the current LEXam leaderboard (36 models)
• Validated all generated JSON files with every_eval_ever validate
• uv run pytest tests/test_lexam_adapter.py -v
• uv run ruff check on changed files

[JoelNiklaus]

Not sure if you want the references to the eval converter in the README or only the general evaluation framework specific ones?