Stability improvements to conversion webhooks

### What problem are you facing?
When I installed provider-aws-kafka v1.0.0 (which contained a v1beta2 resource) into a cluster running crossplane 1.12, [a TLS certificate error](https://github.com/upbound/provider-aws/issues/1147) caused the provider to crash loop, which rendered all the resources it managed basically unusable. The TLS issue was fixed in https://github.com/upbound/provider-aws/pull/1157, and while I have not yet found any other issues that could cause the conversion webhooks to crash, I am concerned that their failure had such a large blast radius.

### How could Upjet help solve your problem?

Could we run the conversion webhooks in such a way that if they panic, they don't take down the entire provider with them? Perhaps in a sidecar container?

Start by locating the conversion webhook entry point and its panic-handling behavior in the Upjet runtime. Determine whether isolating webhook failures is feasible, then define and test behavior that prevents a webhook panic from taking down the provider.

Written by the indexing model from the issue text.