Simplify `FunctionProfilingLayer`

[MegaRedHand]

Our current FunctionProfilingLayer implementation got quite complex, and has some problems: mainly that it requires a String allocation for the namespace (added in #5311), and that implementing functionality via on_enter/on_exit causes it to measure processing time instead of latency.

We should check if it's not easier to use a function wrapper (like record_async_duration), or if these issues can be fixed, and in either case, unify profiling metrics of RPC and block processing.

[rodrigo-o]

I'm doing some test in separating the RPC handling (now accounting also for the result count) from the profiling.rs which was in the beggining just thought to work on block_execution timing. If this is clear and more readable i'll push create a new PR with it, which will left the old profiling as is and easier to refactor in the future if needed.