Improve parallelMap

[kean]

• Remove withoutActuallyEscaping, concurrentPerform already accepts a nonescaping closure
• Avoid sorting results as it might become quite expensive with large N

[mattt]

Thanks for this patch, @kean!

You're totally right about withoutActuallyEscaping being unnecessary here. I was less confident about the correctness of preallocating a fixed-size collection to collect the transformed results (hence my original approach), but after re-reading the Swift Type Layout documentation and checking a few assumptions with MemoryLayout in a Playground, I can't think of any cases where this wouldn't work.

I don't think we need a Changelog entry for this PR since this is an implementation detail, so I'll merge this now.