Handle panics by spawned components in Application::run

[tillrohrmann]

The Application struct spawns different components. If a spawned component panics, this is currently not handled properly. If this happens, then we need to either propagate the panic or fail the process. There are different ways to achieve it:

1. compile the binary with panic = abort to abort on panics
2. Poll the returned JoinHandle to handle panics in spawned components
3. Use catch_unwind to catch panics and handle them (e.g. aborting the process or returning them as errors)

The latter approach is more selective but also prone to errors.

[tillrohrmann]

It is probably the easiest and safest to compile the binary with panic = "abort". That way every panic will lead to a process abort. Additionally, we need to pass force-unwind-tables to the compiler in order to produce helpful stack traces.

[slinkydeveloper]

My 2 cents: Enabling panic = abort is good IMO, as we easily catch mistakes on our side (e.g. bad unwrap()), but I won't rely on it as error handling method for component spawning, as this flag might have unintended side effects we don't know atm. Example: could this affect some dependencies, which might depend on panicking for error handling? Ergo I think it's worth that we also properly handle errors coming out of JoinHandle for this specific case.

[tillrohrmann]

This is a fair comment. If we see that a library uses panics for error handling, then we have to change back from panic = "abort" to explicit handling of panics in spawned tasks.

Explicit handling should be doable by awaiting JoinHandles or using catch_unwind. The reason why I have opted for panic = "abort" is that it is less error-prone (one could easily forget 2. or 3.).