Add VideoToTextPipeline with smart frame sampling and system prompts

[mkorn1]

What does this PR do?

This PR creates a placeholder pipeline in anticipation of the addition of video to text models to the HF ecosystem. In the meantime, the approach is multimodal.

This PR adds a comprehensive VideoToTextPipeline implementation that enables video-to-text generation using image-text-to-text models. The pipeline supports intelligent frame sampling, system prompts for guided generation, and flexible generation parameters.

Key Features

• Smart Frame Sampling: Automatically calculates optimal number of frames based on video duration (1 frame per second, with min 8 and max 128 frames)
• Explicit Frame Control: num_frames parameter allows users to specify exact number of frames to sample
• Time-based Sampling: Uses time-spaced frame selection rather than simple linear sampling for better temporal coverage
• System Prompts: system_prompt parameter enables guided generation with custom instructions
• Flexible Generation: Full support for generate_kwargs to pass any generation parameters to the underlying model
• Batch Processing: Supports processing single videos or batches of videos
• Error Handling: Proper validation for conflicting parameters (e.g., max_new_tokens in both argument and generate_kwargs)

Implementation Details

• Implements VideoToTextPipeline class extending the base Pipeline class
• Uses PyAV for video frame extraction with intelligent temporal sampling
• Supports both local file paths and HTTP/HTTPS URLs for video inputs
• Includes comprehensive logging for debugging frame selection and processing
• Handles GIT models and other image-text-to-text models appropriately

Example Usage

from transformers import pipeline

Basic usage

captioner = pipeline("video-to-text", model="microsoft/git-base")
result = captioner("path/to/video.mp4")

With custom frame count

result = captioner("path/to/video.mp4", num_frames=32)

With system prompt

result = captioner("path/to/video.mp4", system_prompt="Describe this video in detail.")

With generation parameters

result = captioner("path/to/video.mp4", generate_kwargs={"temperature": 0.7, "max_new_tokens": 100})

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).

• Did you read the contributor guideline,
  Pull Request section?

• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.

• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.

• Did you write any new necessary tests?

Who can review?

This PR adds a new pipeline for video-to-text generation, which is a multimodal task. Please review:

• @Rocketknight1 (pipelines)
• @zucchini-nlp (multimodal models, visual-language models)

The implementation follows the existing pipeline patterns and integrates with the existing pipeline registry system.

[zucchini-nlp]

Thanks for the PR @mkorn1 . I just merged any-to-any pipeline (#40884) and I want us to use it for all modalities when possible. Having a separate pipeline for each modality-to-text is not easy to maintain in long-term