Add Google Gemma local LLM support, harden config parsing, improve local LLM handling, and update CI/docs

[AlexTkDev]

Motivation

• Swap the default local model from TinyLlama to Google Gemma 4 to improve offline instruction quality and align docs and defaults.
• Make environment integer parsing robust to invalid or out-of-range values to avoid startup crashes.
• Improve local LLM invocation ergonomics and error messages to provide safer, clearer fallback behavior.

Description

• Add a safe integer parser _get_int_env in config.py and expose LOCAL_LLM_MODEL_PATH, LOCAL_LLM_CONTEXT, LOCAL_LLM_THREADS, and LOCAL_LLM_MAX_TOKENS with sensible defaults.
• Update services/local_llm.py to load the model from LOCAL_LLM_MODEL_PATH, use configured n_ctx/n_threads, normalize max_tokens, harden prompt handling, and provide clearer logging and error strings.
• Adjust services/llm.py import to from services.local_llm import ask_local_llm and preserve optional process-based execution for local LLM calls.
• Update documentation files (README.md, models/README.md) to reflect Google Gemma 4 usage, update model download examples and .env guidance, and change release/version notes to v4.1.0.
• Update CI lint workflow .github/workflows/pylint.yml to use actions/setup-python@v5 and reduce the test matrix to python-version: ["3.10", "3.14"].

Testing

• Added unit tests tests/test_config.py and tests/test_local_llm.py and ran pytest, with the new tests passing.
• Updated CI lint workflow at .github/workflows/pylint.yml to run on push with actions/setup-python@v5 and the updated Python matrix so linting will execute automatically on push.

---

Codex Task