Prevent server crash by restarting child

[prognostikos]

Before this change, if Errno::EINVAL is thrown when sending IO to the child the server would crash. Now we catch the exception and, assuming the child has problems, we start a new child.

[prognostikos]

This is a solution, or maybe better characterized as a work-around, for #716

Before this change, in a Rails 7.0.8.4 app using ruby 3.3.4 (2024-07-09 revision be1089c8ec) [arm64-darwin23] I could trigger the error as follows:

1. start spring using spring server in one terminal
2. start rails c in a second terminal
3. run rails db:migrate:status in a third terminal - it succeeds
4. run touch config/initializers/content_security_policy.rb in the third terminal
5. run rails db:migrate:status in the third terminal - it hangs with no output and the console running in the second terminal is dead.
6. in the first terminal, spring dies with lib/spring/application_manager.rb:64:in `send_io': Invalid argument - sendmsg(2) (Errno::EINVAL)

I have put the spring server logs for both the error and fixed versions here: https://gist.github.com/prognostikos/0d07e599516e51ddc355134e13c94a95

I would love to get this into a new Spring release, what needs to be done for that to happen? I can see there are some issues with CI, and running the acceptance tests locally with RAILS_VERSION="edge" also seems to hang so there is probably a bit of work to do on the test setup.

[prognostikos]

Should we do some kind of cleanup here in case the child is hanging around as a zombie process?

[fschwahn]

This seems to fix the problems I encountered in #716 👍

[fmichaut-diff]

Could this get attention from the Maintainers ? It's a very simple patch that solves a very annoying problem

[ZimbiX]

Thanks for the fix, @prognostikos ❤️ And cheers for the merge, @rafaelfranca! I would love to see a release with this 🙏