Add MemoryCopy IL instruction and take advantage of it in various places

[mkustermann]

We should add a generic MemoryCopy IL instruction, capable of copying ranges of bytes where source and destination can be _OneByteString / _TwoByteString as well as various typed data as long as their element sizes agree.

This can be used for

• Faster UTF-8 decoding
• Faster UTF-8 encoding
• Faster List.setRange() implementation for typed data
• Faster String.fromCharCodes() implementation for Uint8List input
• Faster String.codeUnits implementation

[mkustermann]

@sstrickl is going to look into making TypedData.setRange() faster.

[sstrickl]

Copying comments from Alex's last review of https://dart-review.googlesource.com/c/sdk/+/319521 here so I don't forget to address them later in followup work:

• "Can we do something like CCallInstr to call memmove directly, without going to Dart VM runtime?"
• "We can consider detecting actual cut-off size [of when to switch from using the MemoryCopy instruction to memmove] at run time, depending on the cache sizes."

Also should look at when time allows:

• Generating static call to _checkSetRangeArguments in IL faster than inlining it into setRange prior to calling _fastSetRange or _slowSetRange and passing the calculated count to those methods (either in addition to or instead of end, depending on if they need end).
• Performance regressions on MemoryCopy.{512,4096}.setRange.*.Uint8 on ARM7/ARM8.