Document Upload: Streamline to provider upload

[nishika26]

Describe the current behavior
Currently, document upload is a two-step process which takes a long time:

• POST /documents - Uploads file to S3 and saves to database only
• POST /collections - Internally downloads files from S3, uploads to provider (OpenAI/Bedrock), creates vector store

Describe the enhancement you'd like
Add optional upload_to_provider parameter to document upload endpoin.

New flow would be:

• Upload to S3 (current)
• Save to DB (current)
• NEW: Upload to provider (OpenAI/Bedrock/Gemini) if specified
• Store provider_file_id in document
• Collection creation uses existing provider files → much faster

[nishika26]

Before making a PR for this, we are first researching and experimenting how much time and memory decrease it will give us- https://docs.google.com/document/d/1AhS7sW8OYu_PFNxNgJ5vJ31D5-75td3_mQ-_SOK41OU/edit?tab=t.0