feat(benchmark): support alternative program calls

[e54-bot]

Change

A program-calls check can now list alternatives: [{"call": ..., "args": {...}}, ...] instead of one call/args. Each enabled call node matching any alternative counts once against the shared min/max; overlapping alternatives cannot inflate the count. The existing single-call form, event selection, bounded reads, and unavailable-content behavior remain intact.

The benchmark guide and SPEC-414 document the format. Tests reuse Proximity-repel in OPY and both Workshop locales, and add Workshop fixtures for array removal, filtering with either != operand order, and a negative that includes the event player. Existing single-call tests are unchanged.

Validation

• Full agent benchmark Python suite with the repository-declared publishing dependency installed.
• Focused rule-content tests and defects suite.
• All 12 public benchmark scenarios validate with Wright 0.17.0 and the pinned OverPy 9.7.10 oracle.
• Before/after comparison over real rule content: origin/main raises KeyError('call') for alternatives-only checks; this branch counts each equivalent form once, including overlapping alternatives.
• git diff --check passes. No Rust code changed; Rust gates were not run locally.

Fixes #630