chunkers: build the arm64 hardware AES scan path without -march=+crypto

[ThomasWaldmann]

The bug

The aes-arm64 scan path of the toeplitz-aes / rabin-aes / goldilocks-aes chunkers was gated on __ARM_FEATURE_AES, which is only defined when the whole build targets the crypto extension. Apple's default target does; a Linux or BSD python extension build does not (gcc/clang -march=armv8-a baseline, and setup.py passes no -march). So on Linux arm64:

• the hardware path was never compiled and every build silently ran the portable OpenSSL EVP path (measured 1.5-1.9x slower per scanned byte on an Apple M3; under Linux on the same core: toeplitz-aes 976 MB/s with the fix vs 529 MB/s before),
• BORG_AES_CHUNKER_KERNEL=aes-arm64 was rejected with "not a kernel of this build" while the message still listed it as valid (this is what the RK3328 report in borg2: benchmarking needed #10160 ran into).

The fix

• Enable the crypto extension per function with __attribute__((target("+crypto"))) (gcc >= 6, clang >= 14), the same way the x86-64 path uses target("aes,sse2").
• Decide at run time whether the CPU has the instructions: getauxval(AT_HWCAP) on Linux (as before), elf_aux_info on FreeBSD, always on Apple Silicon; elsewhere only a build that targets +crypto as a whole may assume so (it would not run on a lesser CPU anyway), everything else stays on EVP. NetBSD/OpenBSD detection is left as a TODO.
• A compiler too old for the attribute now reports aes-arm64 as "not compiled into this build" instead of "not a kernel of this build".
• Regression test: on aarch64 the hardware path must be the default whenever the CPU has AES instructions.

Verification

• Linux arm64 (Debian 13 image, gcc 14.2, default flags, __ARM_FEATURE_AES undefined): default kernel is now aes-arm64, BORG_AES_CHUNKER_KERNEL=aes-arm64 is accepted, all chunker tests pass (124 passed), and the hw/evp paths produce identical cut points. With the previous headers in the same build: default evp, aes-arm64 rejected, the new test fails.
• macOS (Apple clang 17): default build unchanged (attribute not needed, PHTE_HW_TARGET empty), chunker tests pass. A kernel-only harness built with -O3 -march=armv8-a+noaes+nosha2 (attribute path) runs at the same speed as the default build and is bit-identical.
• x86-64 (Zen 4, gcc 14): unaffected, vaes default, chunker tests pass.
• FreeBSD arm64 could not be tested here; elf_aux_info, AT_HWCAP and HWCAP_AES come from <sys/auxv.h> (checked against the FreeBSD headers).

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.67%. Comparing base (71dc70e) to head (e46872a).
⚠️ Report is 15 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #10306      +/-   ##
==========================================
+ Coverage   87.56%   87.67%   +0.10%     
==========================================
  Files         103      103              
  Lines       18686    18712      +26     
  Branches     2875     2880       +5     
==========================================
+ Hits        16362    16405      +43     
+ Misses       1622     1606      -16     
+ Partials      702      701       -1     

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