Implement dependency injection for BlipProcessor and BlipForConditionalGeneration in VideoDescriber initialization

[chigwell]

User Story
As a software developer,
I want dependency injection for BlipProcessor and BlipForConditionalGeneration in VideoDescriber initialization
so that I can mock these components during testing and easily swap models for flexibility.

Background
The current implementation directly initializes BlipProcessor and BlipForConditionalGeneration in VideoDescriber.__init__ (see frame_story/video_describer.py lines 13-14). This tight coupling:

• Prevents proper mocking during unit tests (e.g., tests/test_video_describer.py relies on actual model loading)
• Forces unnecessary model downloads during test runs
• Limits customization of model versions/implementations
• Creates longer test execution times due to real model initialization

Acceptance Criteria

• Modify VideoDescriber.__init__ to accept optional processor and model parameters
• Update get_video_descriptions to pass these dependencies through if provided
• Refactor test_video_describer.py tests to use mocked processor/model
• Ensure existing initialization works unchanged when no parameters are passed
• Add type hints for new parameters in method signatures

Validation

1. Verify VideoDescriber() initializes with default Salesforce/blip models when no args passed
2. Confirm tests pass when using mocked processor/model in TestVideoDescriber.setUp()
3. Check that custom models can be injected via VideoDescriber(processor=CustomProcessor(), model=CustomModel())
4. Ensure describe_frames still generates valid descriptions after changes