[llvm] [ARM] Fix VBICimm and VORRimm generation under Big endian. (PR #107813)

[via llvm-commits]

Zhenhang1213 wrote:

> You might well be right if it is casting i32->f64. I hadn't looked at that bit of code, just the code that calls isConstantSplat. Are you able to create a double value that produces incorrect results?

OK， I will try to create it then，if I make any progress，I'll tell you.  This case is right

https://github.com/llvm/llvm-project/pull/107813