Add LLM-based image description to PptxConverter

[keenranger]

This pull request introduces the use of a language model to generate descriptive alt text for images in the markitdown module. The key changes include the addition of a new method to generate descriptions using a language model and the integration of this method into the existing image processing workflow.

Integration of language model for image descriptions:

• src/markitdown/_markitdown.py: Added logic to convert method to use a language model client and model for generating image descriptions if provided in kwargs.

New method for generating descriptions:

• src/markitdown/_markitdown.py: Added _get_llm_description method to generate image descriptions using a language model. This method constructs a data URI for the image and sends a prompt to the language model client to generate the description.

Limitation

BE AWARE BEFORE CONVERTING BIG PPTX FILE
It may take a long time.

Suggestion

• Add a parameter as kwargs, such as “use_llm,” inside the convert method to allow not only PPTX but also other converters to selectively use LLM for dispatching image descriptions.
• Refactor the file handling logic in _get_llm_description into a separate function to make _get_llm_description reusable across all converters.
• Add asynchronous logic for generating image descriptions.

[keenranger]

@microsoft-github-policy-service agree company="Asleep Inc."

[l-lumin]

hi, could you add tests

[keenranger]

@l-lumin Simple LLM test added using the current test.pptx file. I included one page with some basic images.