[llvm] [LangRef] Do not make align imply dereferenceability (PR #158062)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> For codegen, @arsenm can shime in. I'm not aware of any test case in trunk that takes advantage of the alignment.
> 

I believe we have some load widening that relies on this. Not sure if it was in AMDGPU or in the generic DAG combiner. We kind of need the alignment there as a hack, as we don't have properly propagated dereferenceable information. i.e., the flag on the MachineMemOperand is a boolean isDereferencable. 

https://github.com/llvm/llvm-project/pull/158062