Pyfory: Drop‑in replacement serialization for pickle/cloudpickle — faster, smaller, safer

Pyfory is the Python implementation of Apache Fory™ — a versatile serialization framework.

It works as a drop‑in replacement for pickle**/**cloudpickle, but with major upgrades:

• Features: Circular/shared reference support, protocol‑5 zero‑copy buffers for huge NumPy arrays and Pandas DataFrames.
• Advanced hooks: Full support for custom class serialization via __reduce__, __reduce_ex__, and __getstate__.
• Data size: ~25% smaller than pickle, and 2–4× smaller than cloudpickle when serializing local functions/classes.
• Compatibility: Pure Python mode for dynamic objects (functions, lambdas, local classes), or cross‑language mode to share data with Java, Go, Rust, C++, JS.
• Security: Strict mode to block untrusted types, or fine‑grained DeserializationPolicy for controlled loading.