[llvm] [ARM] Stop gluing 1-bit shifts (PR #116547)

[Sergei Barannikov via llvm-commits]

s-barannikov wrote:

@statham-arm I've updated the description, PTAL.

> But I don't see any examples in the changed tests where that's actually happened. All the test changes look as if they've just perturbed the register allocation.

This is probably due to 1-bit shift optimization not triggering often on existing tests. In the next patch I change FP comparisons to use data flow too, and there is a significant difference. Here is just a small example:
![image](https://github.com/user-attachments/assets/ff7bc434-5f86-4906-81cb-ee4ed3675cd4)

I didn't submit it yet because it depends on this PR.


https://github.com/llvm/llvm-project/pull/116547