[BFCL] Retire Executable Categories from Leaderboard

[HuanzhiMao]

BFCL V1 introduced several “executable” categories (rest, exec_simple, exec_multiple, exec_parallel, exec_parallel_multiple) that rely on external APIs. However, these categories have drawbacks that frequently arise in community feedback:

1. Unreliable API availability: The external API can fail randomly, resulting in frequent sanity-check failures and inconsistent evaluation results.
2. Integration difficulties: Foundation model providers struggle to incorporate executable categories into their training pipelines due to the complexity involved.
3. API key management: While the APIs offer free plans, these often lack sufficient bandwidth for evaluating multiple models together, forcing users to upgrade to paid plans.
4. Slow performance: External API calls add latency, slowing down the evaluation process.
5. AST sufficiency: AST-based checks already serve as a robust proxy for executability, reducing the need for external APIs.

Because of these issues, this PR removes all executable categories from BFCL and excludes them from the overall leaderboard accuracy calculation.

[Fanjia-Yan]

LGTM