Avoid using JVM HashCode on Http 2 DefaultStream

[franz1981]

Motivation:

relying on DefaultStream's hashCode forces calling into JVM's strategy for identity hash code

Modifications:

Use the stream's id to identify DefaultStream

Result:

Faster active stream insertion, saving producing JVM's hashcode

[franz1981]

This has been detected on the http 2 hello world-like benchmarking, see (in violet)

In case the stream's id is not good enough, given that DefaultHttp2Connection::createStream is single-threaded, I can use an instance int/long counter per each produced DefaultStream instance and use that one, instead.

[chrisvest]

The stream id is a peer controlled value we read off the network. I'm hesitant to use it here, in case it opens us op to session hijacking vulnerabilities. I'd rather we pulled a hashcode off an atomic counter, and retained object identity for equality.

[franz1981]

@chrisvest looking at the code paths where it is used, it shouldn't happen (because we check upfront the expected next id to create, and removal happen from within the stream itself - which means that the creation id check had worked out), said that, better be safe then sorry, and we can skip using an atomic counter here, given that the default stream code path is single threaded (you can verify it by checking the stream id check before, which uses thread unsafe fields)

[franz1981]

Just fyi, @chrisvest I'm recently receiving hints from users which wrongly compare http 1.1 Vs 2 (very wrongly, cause the power of http 2 is beyond hello world use cases), but effectively there are few low hanging fruits (and some not so low) which could reduce a lot the stream management cost (creation and notification), plus other improvements related known header name lookups (addresses elsewhere, in another pr), and I am trusty, we can get very similar performance by fixing everything, assuming code won't suffer too much from this effort.

[franz1981]

@chrisvest I've added a commit which is leaving equals as it is and allowing ConnectionStream (which extends DefaultStream, but is not placed in the same linked hash set) to behave as it is now.

[normanmaurer]

@franz1981 thanks a lot!