[RISCV] Correct alignment of one-active (de)interleave cases (#150052)
Noticed this while going to rewrite the load case as a DAG combine. I
don't have a test case which demonstrates this leading to a miscompile,
but it seems like it could be possible.