Add codegen-llvm regression tests

[okaneco]

Most of these regressions deal with elimination of panics and bounds checks that were fixed upstream by LLVM.

closes #141497
closes #131162
closes #129583
closes #110971
closes #91109
closes #80075
closes #74917
closes #71997
closes #71257
closes #59352

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Mark-Simulacrum]

This feels a little brittle -- maybe we should check a more specific sequence of instructions? Or some other form of assertion that's not "no adds".

[okaneco]

Sure, I can change this to the specific sequence of checking and-cmp-ret in addition to the check-not.

This was the IR change from that issue, the subtraction being unwanted.

define i1 @src(i8 range(i8 0, 4) %arg) {
-  %add = add nsw i8 %arg, -1
  %and = and i8 %add, -3
  %cmp = icmp eq i8 %and, 0
  ret i1 %cmp
}

[Mark-Simulacrum]

This also seems brittle -- I could see a future compiler making this just call Vec::clear (given that the code sequence is the same for all 4-byte vec values, and maybe even for all T: Copy vec values?).

[okaneco]

I'm not sure what to do so I'll back this test out then.

1.29 codegen - https://rust.godbolt.org/z/9qcPPMh67
1.30 codegen - https://rust.godbolt.org/z/7qK1rojYG

[okaneco]

I removed the vec clear test and that issue from the PR message.
I made the other test an explicit sequence instead of no additions.

@rustbot ready