[AMDGPU] Lower uniform usubsat to SOP (#203155)

Prefer scalar (SALU) lowering for uniform `usubsat`, since usubsat(a, b)
= max(a, b) - b.
* i32: add a GCNPat matching uniform `usubsat` to S_MAX_U32 + S_SUB_I32
* i16: route uniform `usubsat` through `promoteUniformOpToI32` instead
of a TableGen pattern that hard-codes the 0xffff masks. This exposes the
zero-extends as real DAG nodes so KnownBits can fold the masks when the
high bits are already known zero; the promoted i32 usubsat then reuses
the scalar pattern. Promote-and-truncate is safe for usubsat because the
result always fits in the narrow type (unlike uaddsat).

Register USUBSAT with `setTargetDAGCombine` and the promotion dispatch,
return ZERO_EXTEND in `getExtOpcodeForPromotedOp`, and add it to
`isNarrowingProfitable` so divergent i16/i32 keep their native VALU
clamp form.

Co-authored by: Jeffrey Byrnes