Setup benchmarks for all transports

[whyoleg]

Motivation:

Better understand current state and be able to compare new/old/rsocket-java versions on different transpots

Result:

Here are some results based on this run.

Note, results should be compared only from single job, and not between them - e.g it's not possible to compare linux vs macOS results as the runners are different

• K/N comparing to jvm is slower by 3-10 times depending on transport/operation - the biggest difference is when we do requests in parallel (expected)
• local transport
  • multiplexed works better than sequential for RS/RC operations, especially in parallel (expected)
  • sequential works better than multiplexed for RR operations (expected)
• ktor-tcp transport - up to 10x improvement comparing to 0.16.0. RS/RC operations (parallel case) are now faster than java, at least in benchmarks
• new netty-tcp transport - has similar performance to ktor-tcp
• new netty-quic transport - very slow... need to be investigated

[github-actions]

Test Results

  961 files  ±0    961 suites  ±0   2h 18m 38s ⏱️ + 1m 6s
4 355 tests ±0  3 906 ✅ ±0  449 💤 ±0  0 ❌ ±0 
7 973 runs  ±0  7 101 ✅ ±0  872 💤 ±0  0 ❌ ±0 

Results for commit 10720eb. ± Comparison against base commit 33a6735.