=== warp_had_32 / exl3_had_shuffle_32 unification (#1861) ===
bodies whitespace-normalised sha256: 769bac6c2cef9199 (both) -- byte-identical before removal

ptxas sm_89, all instantiations of exl3_a_prep.cu:
  BEFORE: 162 instantiations, regs 30-63, 0 bytes spill in every one, 0 over the 64-reg cap
  AFTER : 162 instantiations, regs 30-63, 0 bytes spill in every one, 0 over the 64-reg cap
  per-instantiation register changes: 0

bit-identity, 8 entry points x 5 in_features (128/2048/5120/6144/17408):
  entries compared 40, elements compared 27,734,202, entries differing 0 -> BIT-IDENTICAL

tests: 97 passed (21 prologue + shipped a_prep/kmajor_layout)
surviving copies of the butterfly loop: 1 (exl3_had128.cuh only)
