Implement a remote caching solution for circuit compilation

[rpanic]

Since in a production setup, we will have a large set of workers, we want them to be able to access a cache such that they don't have to each compile all the circuits every time they get spun up.
Currently, o1js only allows for synchronous caching (with a implementation for storage in the file system). However, we can build a mechanism around it to on-demand fetch artifacts from a S3 bucket to the store on the file system, so that o1js can then read it from the file system.