JIT: Clean up RefPosition "killed registers" storage

[jakobbotsch]

Avoid some ifdefs and rename the field to reflect more specifically what it represents.

Also avoid calling getWeight entirely for RefTypeKill ref positions, even for dumping. Dumping this info does not make much sense since there was a requirement that it was never called in non-dump paths anyway.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[jakobbotsch]

No asm diffs. Some large Linux-hosted JIT TP regressions that I need to look into. The Windows-hosted JIT has negligible TP regressions.

[jakobbotsch]

Looks like #103573 will fix the JITDUMP crash as a separate change.

[jakobbotsch]

The linux-hosted JIT regressions are just the LTO mismatch when PGO is disabled that is being discussed in #103058:

Base: 50016445039, Diff: 50115771931, +0.1986%

84928137 : +53461.67% : 74.74% : +0.1698% : _ZNK9regMaskTP14IsRegNumInMaskE15_regNumber_enum               
20940077 : +8.07%     : 18.43% : +0.0419% : _ZN10LinearScan12processKillsEP11RefPosition                   
309229   : +1.64%     : 0.27%  : +0.0006% : _ZN10LinearScan24allocateRegistersMinimalEv                    
297458   : +0.33%     : 0.26%  : +0.0006% : _ZN10LinearScan25buildKillPositionsForNodeEP7GenTreej9regMaskTP
-7146243 : -0.65%     : 6.29%  : -0.0143% : _ZN10LinearScan17allocateRegistersEv                           

[jakobbotsch]

/azp run runtime-coreclr superpmi-diffs

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jakobbotsch]

cc @dotnet/jit-contrib PTAL @kunalspathak

No asm diffs. linux hosted regressions due to what I mentioned above. Some minor MinOpts TP regressions (up to 0.04% in the contexts with many MinOpts contexts). I think it's worth it to avoid the ifdefs and avoid bugs like the one fixed by #103573.

[kunalspathak]

LGTM. Thanks for cleaning it up. The new name killedRegisters is much better.