perf: Remove python coroutine/trampoline overhead in awaits for ~22% faster dependencies goal

[tobni]

 ➜  ✗ hyperfine --warmup 1 --command-name 'branch' 'PYENV_VERSION=pants@3.14.3                                     
  PANTS_SOURCE=<redacted>/code/pants pants --no-pantsd dependencies ::' --command-name 'main' 'PYENV_VERSION=pants@3.14.3 PANTS_SOURCE=/tmp/pants-main pants --no-pantsd dependencies ::'    
  Benchmark 1: branch                                                                                                                                                                          
    Time (mean ± σ):     25.094 s ±  0.306 s    [User: 30.610 s, System: 8.516 s]                                                                                                              
    Range (min … max):   24.775 s … 25.564 s    10 runs                                                                                                                                        
                                                                                                                                                                                               
  Benchmark 2: main                                                                                                                                                                            
    Time (mean ± σ):     32.304 s ±  0.183 s    [User: 37.097 s, System: 11.844 s]                                                                                                             
    Range (min … max):   32.008 s … 32.614 s    10 runs                                                                                                                                        
                                                                                                                                                                                               
  Summary                                                                                                                                                                                      
    branch ran                                                                                                                                                                                 
      1.29 ± 0.02 times faster than main                                                                                                                                                       

The numbers are measured targeting my company repo.

The improvements scales with the number of rule calls of a goal, as it it pure "per-call" python overhead that is eliminated.

Edit:
I purposefully have changed the typing to signal that the engine selectors aren't coroutines. I lived with the belief they were and that it was important. They were only coroutines because of wrappers, and nothing used the fact (anymore).

[cburroughs]

🤩 I'm just going to note here that _rule_call_trampoline is ~20% of Python on-cpu time in #23236, #18911, and my internal pants check case I mentioned on the last Developer Meeting.

[benjyw]

Tricky to review, but I had looked at this code a few months back for the call-by-name work, so I think this is sound. And if any part of it was not, we'd have seen that in CI.

[benjyw]

Excellent perf gain!