Refactoring fixit sweep, isel through AsmPrinter about llvm-mos

[johnwbyrd]

This is a good idea. But it seems to be more like a principle of ongoing development, as opposed to a specific deliverable. Because refactoring is, and ever will be, both necessary and ongoing, I'm closing this particular issue. If particular things need to be refactored, they should pop up again as new issues.

[mysterymath]

While this is something that should happen in an ongoing fashion, too many of these items have piled up in my half of the codebase to fix incrementally at the rate we're accumulating new code. This issue is for a specific "fixit sweep" through the codebase, starting from the instruction selector (where it left off) and moving through to the end of the AsmPrinter. At the end of this, I'd expect a significant LOC reduction in the code generator, which should make successive tasks quicker.

[johnwbyrd]

When that happens, the AsmParser really ought to be rewritten. It works, but it's unnecessarily verbose due to its AVR heritage, and could probably be done in far fewer lines of code.