[llvm] MachineFunction: Remove null check on TargetRegisterInfo (PR #128480)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> We've not used pure functions here before - are we happy with setting this precedent, and are there any other TargetSubtargetInfo functions that should be made pure?

TargetInstrInfo / TargetRegisterInfo use an inconsistent mix of pure and cases with default implementations which are just llvm_unreachable. I don't think it matters, or why TargetSubtargetInfo would be any different.

In this cas, I don't know why these aren't just required fields in the base class the subclass constructor is required to set 

https://github.com/llvm/llvm-project/pull/128480