code_* cleanup, and strip IR metadata from code_llvm output

[ihnorton]

The first commit cleans up the code_native and code_llvm pathways a little bit by removing an intermediate function and calling the one we want directly from Julia code.

The second commit changes the behavior of code_llvm to strip all metadata before printing the IR. This reduces the work needed to run code_llvm output with lli for debugging purposes. I also added a code_llvm_raw version with the current behavior. If people don't think stripped should be the default I'll switch it around.

before:

; Function Attrs: uwtable
define void @julia_bf_112785(i32, i64) #0 {
top:
  call void @llvm.dbg.value(metadata i32 %0, i64 0, metadata !11, metadata !17)
  call void @llvm.dbg.value(metadata i64 %1, i64 0, metadata !18, metadata !17)
  %sext = shl i32 %0, 24, !dbg !19
  %2 = ashr exact i32 %sext, 24, !dbg !19
  %3 = icmp eq i32 %2, %0, !dbg !19
  br i1 %3, label %pass, label %fail, !dbg !19
...

after:

; Function Attrs: uwtable
define void @julia_bf_112785(i32, i64) #-1 {
top:
  call void @llvm.dbg.value(metadata i32 %0, i64 0, metadata !0, metadata !13)
  call void @llvm.dbg.value(metadata i64 %1, i64 0, metadata !14, metadata !13)
  %sext = shl i32 %0, 24
  %2 = ashr exact i32 %sext, 24
  %3 = icmp eq i32 %2, %0
  br i1 %3, label %pass, label %fail

[ihnorton]

I should probably get rid of the calls to @llvm.dbg.value too, although those are easier to remove as the whole line can be deleted.

[ihnorton]

Force-pushed my fix 3 times, branch never updated, now I hit this bug which I see Arch hit recently too.

😡

[ihnorton]

And apparently this is what happens when trying to create a new PR from the same branch:

[ihnorton]

I seem to have deeply angered the github gods as I can't open a PR for a branch with a different name either.

[tkelman]

blame it on the china ddos? if that's still happening...