Parallel vectorized Keccak-256/SHA-256 merkle tree hashing

[edubart]

This PR:

• Adds custom Keccak-256/SHA-256 implementation using SIMD instructions optimized for hash operations we use in our hash tree (with fixed size leaves/nodes), it uses compiler vector extensions (GCC/Clang compatible) for a simpler implementation (without compiler intrinsics).
• Adds specialized code paths leveraging AVX2 and AVX512 instructions on x86_64 using GCC function multiversioning
• Optimizes pristine page checking with better SIMD instruction utilization
• Makes hash function configurable (adapted from Add configurable hash tree target support #327) - defaults to Keccak-256, can be set to SHA-256 via --hash-tree=hash_function:sha256
• Adds new C API hash functions (cm_get_hash and cm_get_concat_hash)
• Adds Lua hash functions cartesi.keccak256 (renamed from cartesi.kecccak) and cartesi.sha256 using the new C APIs
• Optimizes hash tree multithreading by reducing atomic operations and improving scheduling

The overall speed boost of periodic sparse hashing at 32K cycles relative to base branch (refactor/new-merkle-tree) is more than 2x in average for must benchmark I run (on my hardware supporting only AVX2 instructions).

[edubart]

this was superseded by #333