JIT: Recompute test block weights after loop inversion

[amanasifkhalid]

Fixes #112196.

[amanasifkhalid]

/azp run runtime-coreclr libraries-pgo

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[dotnet-policy-service]

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

[amanasifkhalid]

Diffs show more churn than I expected. I'm seeing quite a few OSR methods where loop inversion changed flow into the OSR entry block, but profile consistency checks didn't flag anything because OSR blocks are excluded from analysis. We now seem to be getting it right in more cases; for example, in LinqBenchmarks:GroupBy00LinqMethodX(), we go from this:

BB19 [0023]  2       BB17,BB18           167.07 167 [039..03D)-> BB20(1)                 (always)                     i IBC hascall gcsafe bwd
BB20 [0004]  1       BB19                165.39 165 [03D..045)-> BB05(0.994),BB21(0.00595)   ( cond )                     i IBC bwd bwd-src osr-entry

to this:

BB19 [0023]  2       BB17,BB18           167.07 167 [039..03D)-> BB20(1)                 (always)                     i IBC hascall gcsafe bwd
BB20 [0004]  1       BB19                167.07 167 [03D..045)-> BB05(0.994),BB21(0.00595)   ( cond )                     i IBC bwd bwd-src osr-entry

[amanasifkhalid]

cc @dotnet/jit-contrib, @AndyAyersMS PTAL. libraries-pgo isn't hitting profile consistency asserts anymore. Looks like SPMI hit an infra issue on win-x64. Thanks!