Allow specialization of the RouterResponder

[Joannis]

I've ran the PerformanceTest codebase two times, before and after this PR. Note that the PerformanceTest only uses 4 cores for HB2 - so that we have resource budget for regular OS use and the wrk client benckmark suite.

Since I made a lot of symbols inlinable, I had to make them public or internal. Public with SPI does not work with inlinable. We'll need to decide on a public API surface if we want benchmarks.

TODOs:

• Fix the router benchmarks in the Benchmarks folder.

Before

Running 15s test @ http://localhost:8080
  8 threads and 128 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.78ms    1.35ms  47.83ms   91.64%
    Req/Sec     9.80k     1.47k   26.85k    89.11%
  1173503 requests in 15.10s, 166.75MB read
Requests/sec:  77697.81
Transfer/sec:     11.04MB

After

Running 15s test @ http://localhost:8080
  8 threads and 128 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.69ms    1.37ms  50.63ms   93.09%
    Req/Sec    10.28k   694.90    13.73k    76.45%
  1233179 requests in 15.10s, 175.23MB read
Requests/sec:  81659.66
Transfer/sec:     11.60MB

[adam-fowler]

Can you get the Benchmarks and Tests working again, to indicate what we would need to make public. I guess for the tests it doesn't matter too much as you can import as @testable.

[adam-fowler]

One comment above

[adam-fowler]

One issue, that was inherited from the original array of structs PR

[adam-fowler]

This shouldn't need to be public. It is only ever used to build a router. I'm not sure why the TrieRouter serialize functions have been made @inlineable. They don't need to be. Building a router doesn't need to be optmized., It is only the resolve functions that should be @inlineable. By removing the inlineable on the serialize functions this can be make internal again.

[adam-fowler]

One minor compile error I fixed, but otherwise looks good