Re-import mlkem-native and mldsa-native; drop custom meta headers

[hanno-becker]

This re-imports mlkem-native and mldsa-native from their respective main branches (mlkem-native 61c831345, mldsa-native 9a180a96b).

Meta headers

This PR removes the three AWS-LC-owned "shadow" backend meta headers: ml_kem/mlkem_aarch64_meta.h, ml_dsa/mldsa_aarch64_meta.h, and ml_dsa/mldsa_x86_64_meta.h. The two AArch64 shadows (added in #3353)
duplicated the vendored meta.h to add a runtime NEON gate with C fallback (reachable via OPENSSL_armcap=0); upstream now provides this gate itself (MLK_SYS_CAP_NEON / MLD_SYS_CAP_AARCH64_NEON), making the
shadows exact duplicates. All backends consume the vendored meta.h as-is.

Full x86_64 ML-DSA backend

The x86_64 mldsa shadow trimmed the vendored meta.h to the assembly-backed subset that AWS-LC imported. The remaining AVX2 C-intrinsic operations have since been rewritten and proved in assembly upstream, so we
now import the full x86_64 backend.

Assembly local labels

Earlier revisions of this PR rewrote mldsa's local assembly labels to Lmldsa_* at import time, since identical labels in mlkem-native and mldsa-native (e.g. Lntt_layer123_start) made the FIPS delocator panic.
Upstream now namespaces local labels per project (pq-code-package/mlkem-native#1813: Lmlk_*; pq-code-package/mldsa-native#1310: Lmld_*), so the workaround is dropped.

Filenames

Upstream has renamed backend .S files to have globally unique names -- e.g. removing clashes between ntt_aarch64_asm.S in both mlkem-native and mldsa-native -- which broke the import into s2n-tls.

Compatibility layer

• Capability func: the custom configs revert to the vendored enum type (mldsa's AVX2 enumerator is now MLD_SYS_CAP_X86_64_AVX2), still mapped to AWS-LC's CRYPTO_is_{AVX2,NEON}_capable().
• ML-DSA API: upstream dropped the sig_len argument from sign/verify (signatures are fixed-length); the ml_dsa.c wrappers are adjusted accordingly. The wrappers now also validate mu_len before calling mldsa-native, which expects a fixed 64-byte mu buffer.
• The mldsa-native update also brings upstream hardening: keygen and signing outputs are zeroized on failure.

Build

• The mlkem assembly lists in crypto/fipsmodule/CMakeLists.txt switch to file(GLOB ... CONFIGURE_DEPENDS), matching mldsa, to absorb upstream .S renames.
• Both importer.sh scripts make GITHUB_SERVER_URL overridable for imports from local mirrors or forks.

Verification

All imported assembly has HOL Light functional-correctness and safety proofs upstream. All assembly except the rejection samplers additionally has proofs of secret-independent timing (the samplers are constant-time as well, but not yet proved so).

---

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license and the ISC license.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 88.00000% with 27 lines in your changes missing coverage. Please review.
✅ Project coverage is 78.47%. Comparing base (6e339a1) to head (bd45929).

Files with missing lines	Patch %	Lines
crypto/fipsmodule/ml_dsa/mldsa/sign.c	51.78%	27 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3367      +/-   ##
==========================================
+ Coverage   78.41%   78.47%   +0.05%     
==========================================
  Files         695      696       +1     
  Lines      124223   124348     +125     
  Branches    17267    17303      +36     
==========================================
+ Hits        97413    97584     +171     
+ Misses      25886    25840      -46     
  Partials      924      924              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jakemas]

Reviewed, this covers the same ML-DSA x86 re-import and sig_len handling as my #3357 (which it supersedes, and with a cleaner verify guard), and since it also brings mlkem-native along I'm happy to go with this PR instead.

[hanno-becker]

CI legitimately failing because the code tries to use the PPC64LE backend which is not imported.

[hanno-becker]

mldsa-native v2 is imminent, so we should hold back merge until this PR is updated.

[jakemas]

mldsa-native v2 is imminent, so we should hold back merge until this PR is updated.

Looks like v2 tasks are now complete, shall we do a fresh import here. I am happy otherwise.

[rod-chapman]

I think this PR needs to be updated to pull the v2.0.0 tags/commits of mldsa-native and mlkem-native?