ast10x0-hace: eliminate remaining panic_is_possible in digest/HMAC path

The no_panics_test scans the compiled ELF and fails if any panic path is
reachable. Two root causes remained in the SHA-2 / HMAC code linked into
the hace_sha256 binary:

- div_by_zero: chunks_exact(4).zip(..) in load_iv / digest_from_context.
  ChunksExact stores its chunk size as a runtime field, so Zip::new emits
  a len / chunk_size division the optimizer cannot prove non-zero.
  Replaced with index-stride loops over length-proven [u8; 4] arrays.

- copy_from_slice len_mismatch_fail: the staging copies in
  HaceDigest::update and HaceHmacCtx::update copy between two range-sliced
  &[u8] of equal-by-construction length the optimizer cannot prove equal.
  Replaced with zip element-wise copies (no length assert, no division).

Also hardened the AES IV copies (AesCipher::crypt, AesSkin CBC chaining)
to use <[u8; AES_BLOCK]>::try_from array assignment instead of
copy_from_slice, per the project no-panic patterns.

no_panics_test now PASSES; nm shows no panic_is_possible, len_mismatch_fail,
or div_by_zero symbols.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>