fix: prevent Groq message mutation in structured output fallback (#562) (#591)

* test: add failing tests for Groq message mutation bug (#562)

Add unit tests and E2E tests that detect the bug where the Groq
structured output code path in llm_invoke.py mutates the shared
formatted_messages list in-place, corrupting fallback model messages.

- 5 unit tests covering both mutation vectors (list .insert() and
  dict content overwrite) plus cumulative corruption
- 3 E2E tests exercising real CSV loading and model selection with
  only litellm.completion mocked

All 8 tests fail on current code and will pass once the fix
(copy.deepcopy at line 1966) is applied.

Fixes #562

* fix: deep copy formatted_messages to prevent Groq structured output mutation corrupting fallback models (#562)

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

* test: trim Groq mutation tests to 3 focused regressions in test_llm_invoke.py (#562)

Replace 770 lines across 2 standalone test files with 3 focused regression
tests (~80 lines) in the existing test_llm_invoke.py, per PR #591 review.

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

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: Serhan-Asad