[Mips][MSA] Lower vector fabs to BCLRI (#228532)
FMAX_A selects an original operand by magnitude and preserves its sign,
so fmax_a(x, x) does not implement fabs. Expand FABS_W and FABS_D to
clear bit 31 and bit 63 with BCLRI, and update their scheduling to match
the bit-clear instructions.
Fixes #228384.
