[clang] [llvm] [InstCombine] Strip leading zero indices from GEP (PR #155415)

[Jan Patrick Lehr via cfe-commits]

jplehr wrote:

I checked out `main` at 69d0c3e44ff yesterday.
Then reverted this commit and validated that this still has the tests pass. I produced "good" intermediate files with that version.
I then reverted the revert and validated that this has the tests fail. I produced "bad" intermediate files with that version.

I have attached all IR files (except precodegen), but for more insight at what exact points in the pipeline they are produced @jhuber6 needs to help me. These are produced with `-Wl,-save-temps` on the final step of the `ninja --verbose` to build that particular test; if that helps in figuring things out.
There are a few differences in the preopt IR files between good and bad. I cannot judge their significance though. Similar for the internalize files.
If there is something else that I can/should provide to help with this, please let me know. I will need guidance then.

[01-preopt-bad.txt](https://github.com/user-attachments/files/22168949/01-preopt-bad.txt)
[01-preopt-good.txt](https://github.com/user-attachments/files/22168950/01-preopt-good.txt)
[02-internalize-bad.txt](https://github.com/user-attachments/files/22168952/02-internalize-bad.txt)
[02-internalize-good.txt](https://github.com/user-attachments/files/22168953/02-internalize-good.txt)
[03-opt-bad.txt](https://github.com/user-attachments/files/22168955/03-opt-bad.txt)
[03-opt-good.txt](https://github.com/user-attachments/files/22168957/03-opt-good.txt)

https://github.com/llvm/llvm-project/pull/155415