Reduce use of self-hosted runner and use cache

[Kailai-Wang]

resolves #259
resolves #393

This PR:

• uses github runner when possible in GHA, currently our self-hosted runner is used to build docker image only, as it's significantly faster than the github runner
• uses cargo-chef + buildx local cache to cache docker layers for regular CI, for more experiments refer to Accelerate the rust compilation #259
• use sccache to accelerate the native binary compilation
• use path-filter to optionally skip the docker building process (e.g. when dependabot bumps the ts-tests dependencies)
• separate dev and prod Dockerfile
• upgrade the upload/download action to v3

Overall it was an interesting research and I've learned new things from the experiments.
I'd like to keep the caching mechanism like this and see how it goes, we could revert it anytime if we think it doesn't fit well.