### `wai_pq_vectors` runs on Windows' main-thread stack

- **`pq_sig::HybridKey`** holds its expanded ML-DSA-65 key (about 47 KiB) on the heap. Held by
  value, each key took that much of every frame that made or moved one in a build without
  optimisation, and `wai_pq_vectors verify` needed about 3.4 MiB of main-thread stack, past the
  1 MiB Windows gives a program: it stopped there with a stack overflow. It now needs about
  0.75 MiB, most of it ML-DSA's own key generation. The keys, the signatures and the corpus are
  the same bytes.
- **`wai-conformance.yml`**: the Windows leg's corpus step no longer links its tools with an
  8 MiB main-thread stack. Each runs on the stack a user's build of it gets, so a tool that
  outgrows Windows' 1 MiB fails there.