2.1.0

Changed

• Improved rpack.pack() search behavior for thin-rectangle pathological
  inputs by adding staged coarse candidate stepping with bounded local
  refinement.
• Documented the thin-rectangle performance benchmark results in the
  documentation.
• rpack.pack() now retries overflowing inputs via bigint preprocessing:
  exact axis-wise gcd normalization first, then conservative power-of-two
  scaling when needed so very large Python integers can still be packed.
• Public helpers rpack.bbox_size(), rpack.overlapping() and
  rpack.packing_density() now transparently handle bigint packings by
  falling back to Python integer arithmetic when C long conversion overflows.

Bugfixes

• Restored the public helper docstrings after wrapping the public helpers for
  bigint fallback support, so help() and the generated API reference stay
  populated.