[Feature] Request panic handler extension point

[yarolegovich]

Details

In the current implementation we recover from panics when AgentExecutor.(Execute|Cancel) fail.

The code runs in a detached context so it's not easy to bubble-up the error as it might be that there are no subscribers who can handle it.

Simple recovery and execution failing might silently hide serious problems, so it would be good to make it into a request handler option like WithExecutionPanicHandler(). This give user flexibility to implement strategies like (for example) panic in test environments and log panics in a specific way in prod.