[serve] catch errors caused by bad deployment configurations

[iasoon]

Why are these changes needed?

When a deployment is faulty due to bad user configuration (such as in #19771, #20396) the serve controller code will error forever. This PR catches deployment errors and reports them back to the deploy caller.

Related issue number

Closes #20396

Checks

• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[edoakes]

@iasoon looks like there's a merge conflict. Also, I don't think we can rely on tblib here as it's a third-party dependency and we want to avoid dependencies to keep ray as slim as possible (unless they're really needed). You can see the ones currently depended on here:
https://github.com/ray-project/ray/blob/master/python/setup.py

[edoakes]

I believe we are currently passing the traceback as a string -- the relevant code is here:
https://github.com/ray-project/ray/blob/master/python/ray/exceptions.py#L66

We should be able to do the same thing here?

[edoakes]

Aside from tblib part, the changes looks reasonable to me

[iasoon]

@edoakes I switched the implementation to use RayTaskError, similar to errors that happen when calling a deployment. Does that look okay to you?

[jiaodong]

Looks good to me as well, thanks for helping to improve this !

[edoakes]

RayTaskError might not be ideal but I can't think of a better alternative for now. Thanks for the contribution :)