[mono][jit] Paired stores and loads are now used for vtypes in regs on arm64

[jandupej]

When loading into memory a vtype stored in multiple GP regs, paired loads are used. Analogously, paired stores are used when possible. This change affects only mini.

static Vector128<float> DUT (Vector128<float> a)
{
   return a + Vector128<float>.Zero;
}

Original codegen:

0000000000000000        stp     x29, x30, [sp, #-0x30]!
0000000000000004        mov     x29, sp
0000000000000008        str     x0, [x29, #0x20]
000000000000000c        str     x1, [x29, #0x28]
0000000000000010        eor.16b v1, v1, v1
0000000000000014        ldr     q0, [x29, #0x20]
0000000000000018        fadd.4s v0, v0, v1
000000000000001c        str     q0, [x29, #0x10]
0000000000000020        ldr     x0, [x29, #0x10]
0000000000000024        ldr     x1, [x29, #0x18]
0000000000000028        mov     sp, x29
000000000000002c        ldp     x29, x30, [sp], #0x30
0000000000000030        ret

After change:

0000000000000000        stp     x29, x30, [sp, #-0x30]!
0000000000000004        mov     x29, sp
0000000000000008        stp     x0, x1, [x29, #0x20]
000000000000000c        eor.16b v1, v1, v1
0000000000000010        ldr     q0, [x29, #0x20]
0000000000000014        fadd.4s v0, v0, v1
0000000000000018        str     q0, [x29, #0x10]
000000000000001c        ldp     x0, x1, [x29, #0x10]
0000000000000020        mov     sp, x29
0000000000000024        ldp     x29, x30, [sp], #0x30
0000000000000028        ret

[jandupej]

/azp run runtime-extra-platforms

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jandupej]

All failures seem to map to known issues. Merging.

[matouskozak]

LGTM!