[llvm] [InstrProf] Change step from 64-bit to pointer-sized int (PR #83239)

[Brian Smith via llvm-commits]

briansmith wrote:

@jdmitrovic-syrmia As an alternative solution, why not implement a (perhaps slow) atomic 64-bit fallback implementation that uses a global lock?

https://github.com/llvm/llvm-project/pull/83239