[AArch64] Optimise MOVI + CMGT to CMGE (#74499)

This fixes a regression that occured for a pattern of MOVI + CMGT
instructions, which can be optimised to CMGE. I.e., when the signed
greater than compare has -1 as an operand, we can rewrite that as a
compare greater equal than 0, which is what CMGE does.

Fixes #61836