Harden allocator correctness and stability

[antono3]

Summary

• add independent 20,000-operation reference models for buddy and ring allocation
• prove synchronized and plain range/buddy allocators produce equivalent results across deterministic traces
• cover zero-capacity state, failure atomicity, and allocation-ID wraparound
• expand benchmark suite v3 with synchronized overhead and 1/2/4/8-worker contention measurements
• retain benchmark output in CI and add cross-platform release baseline artifacts
• document SemVer stability, compiler support, ownership boundaries, complexity, and threading guarantees

Validation

• released V 0.5.2 (7647ce1): format, vet, and public documentation generation
• released V 0.5.2: 8/8 test files and all public examples
• released V 0.5.2: benchmark suite v3 quick workload with trace/checksum equivalence
• released V 0.5.2 + Clang: AddressSanitizer and UndefinedBehaviorSanitizer, 8/8 test files
• released V 0.5.2 + Clang: ThreadSanitizer contention suite
• both GitHub Actions workflow files parsed as YAML