Add Google Gemini as an LLM provider

[Ahmed-Ezzat20]

Summary

• Adds Google Gemini as a third LLM provider alongside OpenAI and Ollama
• Implements _GeminiClient that mimics the client.chat.completions.create() interface, so all existing pipeline stages (describe, review, translate, resegment, thumbnails) work without changes
• Supports text, multimodal (base64 images), and streaming via generate_content_stream
• Converts OpenAI message format to Gemini's Content/Part types (including system → system_instruction)
• New optional dependency: pip install "mazinger[llm-gemini]"

Changes

File	Change
llm.py	New _GeminiChatCompletions, _GeminiChat, _GeminiClient classes; gemini_api_key param on build_client()
cli/_groups.py	New add_gemini() helper, wired into add_llm() and make_llm_client()
cli/_dub.py	Pass gemini_api_key to MazingerDubber
cli/_transcribe.py	Pass gemini_api_key to build_client() for ASR review
pipeline.py	gemini_api_key on __init__ and _llm_client()
pyproject.toml	New llm-gemini optional extra (google-genai>=1.0)

Usage

# Set API key
export GEMINI_API_KEY=your_key

# Full dubbing pipeline with Gemini for all LLM stages
mazinger dub video.mp4 --llm-model gemini-2.5-flash

# Or pass key explicitly
mazinger dub video.mp4 --gemini-api-key KEY --llm-model gemini-2.5-flash

Test plan

• Verified text completion with gemini-2.5-flash — correct response and usage metrics
• Verified multimodal (image) completion — correctly identifies image content
• Verified CLI parser builds without conflicts across all subcommands
• Verified MazingerDubber accepts and passes gemini_api_key through to build_client()
• Test full dub pipeline end-to-end with Gemini as LLM provider