Adjust CMP with immediate to CSE opportunities with other CMP

[vladimirradosavljevic]

For the following test:

declare void @use1()
declare void @use2()
define i256 @test(i256 %arg, i256 %arg1, i256 %arg2, i256 %arg3) {
entry:
  %icmp = icmp ult i256 %arg, 5
  br i1 %icmp, label %bb1, label %bb2

bb1:
  call void @use1()
  br label %bb4

bb2:
  %icmp2 = icmp ugt i256 %arg, 5
  br i1 %icmp2, label %bb3, label %bb4

bb3:
  call void @use2()
  br label %bb4

bb4:
  %phi = phi i256 [ 152, %bb1 ], [ 42, %bb3 ], [ 0, %bb2 ]
  ret i256 %phi
}

Following assembly is generated:

test:                                   ; @test
	sub.s!	4, r1, r0
	jump.gt	@.BB0_2
	near_call	r0, @use1, @DEFAULT_UNWIND
	add	152, r0, r1
	ret
.BB0_2:                                 ; %bb2
	sub.s!	6, r1, r0
	add	r0, r0, r1
	jump.lt	@.BB0_4
	near_call	r0, @use2, @DEFAULT_UNWIND
	add	42, r0, r1
.BB0_4:                                 ; %bb4
	ret

In this case, if we adjust both compares to:

	sub.s!	5, r1, r0
	jump.ge	@.BB0_2
        ...
	sub.s!	5, r1, r0
	jump.le	@.BB0_4

Second sub.s! 5, r1, r0 is not needed, and MachineCSE would eliminate it.

This optimization opportunity is inspired by AArch64ConditionOptimizer, so implement the same optimization for EraVM.