P sync/sync

[AlePouroullis]

Sync operation for the python SDK

Functionality

• Implemented .sync() function on the extended Humanloop client
• The sync() function internally called utils defined in src/humanloop/sync/sync_utils.py (added to .fernignore so that it doesn't get overriden by the auto-generated SDK)
• Composed of existing client functionality, namely list_files() and serialize()
• Supports prompt and agent files
• Only allows for single version per file. Creates identical directory structure to that of the HL workspace (e.g. A/B/C where C is a prompt file will be stored in humanloop/A/B/C.prompt
• Pulls default deployed files
• No controls, like for environment, fetching from specific directory, etc. Only uses defaults: production files for the whole workspace

Tests

• Wrote integration tests that use local environment (the client in the test uses localhost:80/v5 for the backend instead of prod)
• Creates test prompt and agent files, and cleans up after
• Checks that the path of the file on remote exists in the humanloop directory

Implementation details:

• list_files() is called, and paginated to get all files
• After a page is retrieved, each individual file is handed off to a thread pool where a thread processes a file, serializing it and writing it to disk

Caveats:

• The serialize() function from the auto-generated SDK returns None, presumably because its response type is text/prompt on the API spec and fern doesn't know how to handle this. Consequently, it returns HttpResponse(response=_response, data=None) when it should do something like return HttpResponse(response=_response, data=_respnose.text) (which I did in this PR as a dirty workaround). I've flagged on the fern slack channel
• Pagination is not done concurrently. Could be sped up by handing off fetching each page to a thread but difficult to coordinate
• This operation could benefit quite a bit from async, but quite complicated.