Fix differences between bash and cmd versions of crossgen test script

[kg]

Thanks to @davmason for finding this.

[kg]

/azp run runtime-coreclr crossgen2 outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[davmason]

It looks like it will solve the issue we found, I am not an expert on this part of the tests but if the outerloop is in a better place than we should ship it

[kg]

Will address all the PR feedback tomorrow now that the outerloop run finished. It looks like this reduced the failure count on outerloop by 70, but it seems like the comparison lanes are broken and didn't run.

[ivdiazsa]

I left a Nit comment but besides that, @janvorli has pointed out all the observations I also had. So once those are addressed, this lgtm! Thank you so much for fixing this Katelyn!

[kg]

One last PR comment to address. I ran shellcheck over the generated script and fixed a bunch of stuff it found (the rest of the warnings are spurious or are located outside of CrossGen.targets).

[kg]

/azp run runtime-coreclr crossgen2 outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[kg]

Got confused for a bit because Not-Int32 still fails with an invalid RVA even after this fix, but that's a separate problem.

[kg]

Composite linux x64 checked was at 94 failures on my most recent run of it outside this PR, and this PR brings it down to 80 failures by fixing most (not all) of the invalid RVA errors that were caused by errors in the script. Merging.