tests(evaluators): add LLM evaluator coverage gap tests mirroring Luna

Add test_llm_coverage_gaps.py covering all utility helpers, operator
branches, payload preparation, client connection config, pool rotation,
JWT auth, CA file fallback, error propagation, and config validation.
Brings llm/client.py from 54% to 97%, llm/evaluator.py from 65% to 97%,
and llm/config.py from 78% to 98% patch coverage.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: wrisa