feat: Implement ARM CPU feature detection across Linux, Android, BSD, Windows, macOS, and iOS

[Xor-el]

Description:

This PR replaces the stub // TODO implementations in TArmSimdFeatures with real, cross-platform ARM hardware capability detection, enabling runtime SIMD dispatch on ARM targets.

New Units

• HlpArmHwCapProvider — Platform-abstracted ARM HWCAP provider:

  • Linux/Android: Resolves getauxval via dlsym to query AT_HWCAP/AT_HWCAP2
  • BSD (FreeBSD, NetBSD, DragonFlyBSD): Resolves elf_aux_info / _elf_aux_info via dlsym
  • Windows ARM64: Wraps IsProcessorFeaturePresent for PF_ARM_V8_* constants
  • Includes HWCAP bit definitions for both AArch64 and ARM32

• HlpDarwinSysCtl — Darwin-specific (macOS/iOS) feature detection via sysctlbyname, resolved dynamically through dlsym. Supports modern FEAT_* sysctl keys (macOS 12+) with automatic fallback to legacy keys (macOS 11).

Changes to Existing Units

• HlpArmSimdFeatures — All CPUHas* methods now contain real detection logic for NEON, AES, SHA1, SHA256, SHA512, SHA3, PMULL, SVE, and SVE2. Added CPUHasCRC32 / HasCRC32 for CRC32 instruction detection.

• HlpX86SimdFeatures — Renamed FSimdLevel → FActiveSimdLevel and GetSimdLevel() → GetActiveSimdLevel() for consistency and clarity.

• HlpCpuFeatures — TCpuFeatures.X86 and TCpuFeatures.Arm properties are now conditionally compiled under HASHLIB_X86 / HASHLIB_ARM respectively, so only the relevant architecture code is included in each build.

• Include files (HashLib.inc, HashLibFPC.inc):

  • Renamed HASHLIB_ARM (which previously meant ARM32) to HASHLIB_ARM32 and HASHLIB_ARM_ASM to HASHLIB_ARM32_ASM to avoid ambiguity
  • Introduced HASHLIB_X86 (= I386 or X86_64) and HASHLIB_ARM (= ARM32 or AArch64) as architecture-family defines
  • Normalized indentation throughout

• All dispatch units (HlpAdler32Dispatch, HlpCRCDispatch, HlpBlake2BDispatch, HlpBlake2SDispatch, HlpBlake3Dispatch, HlpSHA1Dispatch, HlpSHA2_256Dispatch, HlpSHA2_512Dispatch, HlpSHA3Dispatch, HlpXXHash3Dispatch, HlpArgon2Dispatch, HlpScryptDispatch) — Updated to call GetActiveSimdLevel() instead of GetSimdLevel().

Design Notes

• All platform API resolution uses dlopen(nil) + dlsym at runtime, avoiding hard static imports and gracefully degrading when symbols are unavailable.
• ARM32 vs AArch64 differences in HWCAP register layout (e.g., ARM32 crypto extensions live in HWCAP2 while AArch64 uses HWCAP) are handled via conditional compilation.
• On Windows ARM64, PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE bundles AES, PMULL, SHA1, and SHA256 together per Microsoft's documentation. SHA3/SHA512 use the newer PF_ARM_V82_* constants.