Add StableRadixSort improvements and hybrid_vector container

[LinnJz]

This PR optimizes StableRadixSort and introduces hybrid_vector (with
HybridAlloc) to support its runtime-sized bucket allocations.

StableRadixSort

• Rewrite with LSD/MSD traversal + sequential/parallel execution (via OpenMP),
  offering four combined modes
• Add compile-time configuration options (RadixSortOptions, BitsPerPass, sign
  handling, NaN positioning)
• Unified interface via radixSort()
• Enhanced benchmarks and test coverage

hybrid_vector & HybridAlloc

• hybrid_vector: a fixed-capacity vector with stack-first, heap-fallback
  storage (threshold configurable, default 1024 bytes)
• HybridAlloc: RAII-free stack-or-heap allocation macros
• Motivation: StableRadixSort's OpenMP parallel path needs per-thread bucket
  storage. The number of threads is runtime-known, so stack-only allocation
  (alloca) isn't viable directly. BitsPerPass::Bits8 (256 entries) fits stack;
  Bits16 (65536 entries) requires heap. hybrid_vector provides a unified
  abstraction that auto-selects the backing storage.

Additional

• Extend CPortability.h with platform detection macros
• Extend Iterator.h with new iterator utilities
• Add folly/docs/hybrid_vector.md documentation
• Update CMakeLists.txt and BUCK files for all new targets