feat: Only allow one Chipper call at a time

[awalker4]

Chipper V2 is very memory hungry. While we work to optimize this, we need to restrict the server to one call at a time. While the model is in use, we'll return a 503 "Please try again". Our hosted API should scale up to meet demand, so the next call should route to an available server.

This includes a refactor to how partition_kwargs are passed to either parallel mode, local partition, or local partition with the new Chipper protection.

To verify, try calling Chipper twice:

curl -X POST 'http://localhost:8000/general/v0/general' --form files="@$file" --form strategy=hi_res --form hi_res_model_name=chipper &
curl -X POST 'http://localhost:8000/general/v0/general' --form files="@$file" --form strategy=hi_res --form hi_res_model_name=chipper

The second call will get a 503 response.

Other changes:

• Return a 400 error if Chipper isn't loaded. The model is private, make sure we explain this for users who self host
• Pass the huggingface token to make docker-start-api for better dev experience
• Add a make docker-start-bash while we're in here

[sentry-io[bot]]

Suspect Issues

This pull request was deployed and Sentry observed the following issues:

• ‼️ ValueError: file '<_io.BytesIO object at 0x7fa14d2c9ad0>' is not a Word file, content type is 'application/vn... /general/v0/general View Issue
• ‼️ Server is under heavy load. Please try again later. /general/v0/general View Issue
• ‼️ PSSyntaxError: Invalid dictionary construct: [/'BM', /'Normal', /'CA', 1, /'LC', 0, /'LJ', 0, /'LW', 2, /'ML', 4... /general/v0/general View Issue
• ‼️ TypeError: cannot unpack non-iterable PDFObjRef object /general/v0/general View Issue
• ‼️ PDFSyntaxError: No /Root object! - Is this really a PDF? /general/v0/general View Issue

_{Did you find this useful? React with a 👍 or 👎}