feat: PT-541: rewrite basically the whole thing & add support for modelcache

[ef0xa]

PT-541: runpod-workers/worker-sglang: update for modelcache

Took me ages to figure out how this worked. It's not really that complicated, but it's been quite obscured by the control flow jumping around all over the place. Should be a bit easier to follow now.

To summarize the changes:

• combine all source files into single handler
• actual comments
• add support for huggingface modelcache
• add type hints to the code
• remove unreachable code
• update version of runpod-python client to 1.7.7

[ef0xa]

@pandyamarut : what's the file used for? can we delete it?

[deanq]

This drives the config forms at quick deploy. If you need any env var that User should be able to configure, it goes here. This schema definition is parsed over to our console as form and inputs. This was created for vllm at first to support versions and changes in their configs. We're using it for all quick deploy templates in the future.

[deanq]

Don't do this. ~= covers for 1.7.0-1.7.x. If/when we go 1.8.x it means breaking changes. We don't want this upgrading before we can test it. Revert this line please.

[deanq]

This drives the config forms at quick deploy. If you need any env var that User should be able to configure, it goes here. This schema definition is parsed over to our console as form and inputs. This was created for vllm at first to support versions and changes in their configs. We're using it for all quick deploy templates in the future.