Polish "Add metrics support for Netty 4.x"

[izeye]

This PR polishes the "Add metrics support for Netty 4.x" changes a bit.

See gh-3742

[jonatan-ivanov]

@izeye Thank you for the changes! Could you please fix the issue number in the commit message and in the description? I think it is not the one that this PR polishes.

[izeye]

@jonatan-ivanov Oops, sorry. I fixed them now.

[franz1981]

@jonatan-ivanov what about the allocatorId ? I didn't yet opened an issue but would be better to:

• use a CopyOnWriteArraySet with the refs of allocators already registered to verify a certain allocator isn't observed yet
• and a static atomic long to create runtime fake identity for the allocator just registered

[bclozel]

@franz1981 the current design assumes that there are various ways to instrument Netty (and the allocator) here:

• manually during startup, when resources like allocators are manually set and known. In this case, there would be no need to maintain the list of already registered allocators
• at runtime, lazily, when an allocator is seen during channel init. In this case, it's a good idea to maintain such a list and prevent duplicate instrumentations

We can't choose this for libraries, this is why we chose to remain neutral in Micrometer's instrumentation. Does this make sense?

[franz1981]

We can't choose this for libraries, this is why we chose to remain neutral in Micrometer's instrumentation. Does this make sense?

Yeah, that's fair, but hashCode suffer from collision, hence, although neutral is not a proper identity and can cause 2 different allocator refs to be reported as the same - it's rare, but possible