codegen: fix VU0 dest-mask lane-bit ordering and VMR32 shuffle pattern typos

[jlsandri]

Problem

Two copy-paste typos in the VU0 macro codegen, both producing silently-wrong output for any instruction that uses a non-default dest-mask or the VMR32 rotate.

1. dest_mask lane-bit ordering inverted across ~36 VU0 emitters

Many emitters were passing the dest-mask bits to _mm_set_epi32() as

(dest_mask & 0x8) ? -1 : 0, (dest_mask & 0x4) ? -1 : 0,
(dest_mask & 0x2) ? -1 : 0, (dest_mask & 0x1) ? -1 : 0

and the VSQD emitter additionally had

{0x1, 0x2, 0x2, 0x1}

as a clearly-typo'd variant of the same pattern.

The correct ordering — matching the rest of the VU0 dest-mask code and the PS2 convention where bit 0 is the X lane — is

(dest_mask & 0x1) ? -1 : 0, (dest_mask & 0x2) ? -1 : 0,
(dest_mask & 0x4) ? -1 : 0, (dest_mask & 0x8) ? -1 : 0

Because _mm_set_epi32() takes its arguments in reverse lane order (arg 0 → lane 3, arg 3 → lane 0), the inverted bit order caused the X/Y mask bits to gate the Z/W lanes and vice versa. Any masked write with a non-uniform dest_mask silently misrouted.

Affected emitters include: VSQD, VADD_Field / VSUB_Field / VMUL_Field / VMADD_Field / VMSUB_Field, VMINI_Field / VMAX_Field, VFTOI* / VITOF*, the _bc variants, the _q variants, the _i variants, and assorted VMULA / ADDAq sites. ~36 call sites in total.

2. VU0_S2_VMR32 shuffle pattern

// before
_MM_SHUFFLE(0, 0, 0, 1)
// after
_MM_SHUFFLE(0, 3, 2, 1)

VMR32 is a 1-lane rotate, not a broadcast of lane 1. The old pattern produced {lane1, lane1, lane1, lane1} instead of the correct {lane1, lane2, lane3, lane0}.

Scope

• One file: ps2xRecomp/src/lib/code_generator.cpp
• +68 / -68 (almost all sites are pure re-orderings of an _mm_set_epi32 call)
• Zero behavioural change for dest_mask == 0xF (the most common case); every other mask was silently wrong before.

Rationale

Both classes of bug are pure copy-paste errors. The fix restores the stated invariant (bit 0 ↔ X lane) across every emitter, and corrects the VMR32 shuffle to perform the rotate the mnemonic actually describes.

[jlsandri]

Closing as part of a batch cleanup after #107 landed. The runtime ecosystem refactor in #107 substantially reworked the files this PR touched, and I would like to re-audit the underlying fix against the new code structure before putting it back in front of you. If the fix is still needed after that re-audit, I will re-open as a focused PR rebased onto current main. Thanks for your patience.