[JIT] Reduce FP multiplication by -1 to plain negation

[neon-sunset]

Description

Given operations like x *= -1, it appears JIT currently does not reduce this operation, using costlier FP multiplication.

Clang/LLVM:

double negate_mul(double num) {
    return num * -1;
}

double negate(double num) {
    return -num;
}

.LCPI0_0:
        .quad   0x8000000000000000
        .quad   0x8000000000000000
negate_mul:
        xorps   xmm0, xmmword ptr [rip + .LCPI0_0]
        ret

.LCPI1_0:
        .quad   0x8000000000000000
        .quad   0x8000000000000000
negate:
        xorps   xmm0, xmmword ptr [rip + .LCPI1_0]
        ret

RyuJIT:

static double NegateMul(double num) => num * -1;
static double Negate(double num) => -num;

// coreclr trunk-20251207+86dfeb91ce270da17416069463e33363bda3ec55

Program:.ctor():this (FullOpts):
       ret      

Program:NegateMul(double):double (FullOpts):
       vmulsd   xmm0, xmm0, qword ptr [reloc @RWD00]
       ret      
RWD00  	dq	BFF0000000000000h	;           -1

Program:Negate(double):double (FullOpts):
       vxorps   xmm0, xmm0, xmmword ptr [reloc @RWD00]
       ret      
RWD00  	dq	8000000000000000h, 8000000000000000h

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[tannergooding]

This one should be fine. They are technically not the same for NaN, but that is an allowed difference and is considered "value preserving".

Simplifying 1 * x to just x is also legal, as is 0 + x when x is not itself +/- 0, but I believe we're already covering these ones.