[Feature Request] Add option to cache model compilation for modular/max-openai-api

[remorses]

What is your request?

I tried deploying modular/max-openai-api to fly.io, but it takes a lot of time to do the first compilation of the model, is it possible to cache the model compilation on disk?

What is your motivation for this change?

Add --model-compile-cache=/.root/model parameter

Any other details?

Fly.io is a serverless GPU deployment platform, the machine is stopped and started often, now model compilation is too slow to be able to deploy in this kind of infrastructure

[ehsanmok]

Thanks! we do have a serialization support but isn't documented well which we'll fix and popularize when more testing is done.