Some of these tests have multiple vid sequence build_vectors in them, and machine CSE tidies up their output.
But now that small build_vectors are sometimes lowered to scalar vmv.x.ss, the output is harder to read with no CSE, e.g. see buildvec_no_vid_v4i8.
This patch splits them up into separate functions to address this, and also makes the diff in D157976 more clear since it causes vmv.x.s to be lowered to more frequently.