Rework semantics: explicit empty ranges and honest partial-order support

[adriendellagaspera]

The crate previously returned an arbitrary, representation-dependent answer
for empty ranges, and PR10's partial_rcmp -> Option<RangeOrdering> collapsed
a poset position into a single ambiguous verdict, conflating "incomparable
with the whole range" with "comparable with one bound, incomparable with the
other".

Both issues are addressed with a single core idea: represent a value's
position as the pair of its relationships to the lower and upper bounds.

• Add BoundOrdering (Within / Outside / Incomparable) and RangePosition
  (the lower/upper pair), with RangePosition::ordering() collapsing to a
  simple verdict when both bounds are comparable, and is_inside().
• RangeOrdering gains an Empty variant. rcmp (Ord) now detects empty
  ranges from their bounds and reports Empty instead of a bogus
  Below/Above, and never falsely reports Inside for an empty range.
• Introduce PartialRangeOrd::partial_rcmp -> RangePosition for PartialOrd
  types, preserving per-bound information instead of losing it in an Option.
• Tests: convert the partial suite to the new type, complete the previously
  empty stubs, and add the comparable-to-one-bound-only cases that the old
  design lost. Fix the README/code name drift (RangeComparable/range_cmp).

This is a breaking change; bump to 0.3.0.

https://claude.ai/code/session_01CTKxSkrLTyauoiGSFo2dVS