[llvm] [SDAG] Don't treat ISD::SHL as a uniform binary operator in `ShrinkDemandedOp` (PR #92753)

[Simon Pilgrim via llvm-commits]

================
@@ -1834,9 +1838,30 @@ bool TargetLowering::SimplifyDemandedBits(
 
       // Try shrinking the operation as long as the shift amount will still be
       // in range.
----------------
RKSimon wrote:

This is now very similar to the `(shl i64:x, K) -> (i64 zero_extend (shl (i32 (trunc i64:x)), K))` fold below, this patch is a bugfix so no need to merge them yet, but maybe add a TODO?

https://github.com/llvm/llvm-project/pull/92753