v1.1.0

Additive and backward compatible. plan(...) keeps its signature and
semantics; every 1.0.0 test passes unchanged.

Added

• Planning between sets. CBiRRT.solve(problem) takes a
  PlanningProblem made of a JointSpace, a start set, a goal set, a
  validator, and an optional path-admissible set. plan(...) now lowers its
  arguments into this representation. See docs/design.md.
• State sets (pycbirrt.sets): a membership-only StateSet protocol
  with optional SetSampler, SetDistance, and SetProjector capabilities;
  FiniteSet, PredicateSet, EmptySet; AnyOf (union) and AllOf
  (intersection) with arbitrary nesting and explicit capability rules; the
  named strategies MostViolatedProjection and RejectionSampling;
  supports, is_finite, members.
• TSRConfigurationSet, the configuration-space set a TSR induces
  through forward kinematics, with membership, distance, sampling, and
  IK-based projection. tsr_weights gives the volume-weighted mixture.
• JointSpace: joint limits, angular-aware metric and direction,
  interpolation, and uniform sampling in one object (planner.space).
• Distinct tolerances on CBiRRTConfig: membership_tolerance,
  connection_tolerance, edge_resolution, progress_tolerance,
  projection_progress_tolerance. Defaults reproduce 1.0.0 behavior.
• PlanResult gains start_source and goal_source (which alternative a
  path used, as a path through nested sets), failure_reason,
  planning_time, tree_sizes, and the search trees tree_start and
  tree_goal for inspection.
• abort_fn on CBiRRTConfig to stop planning early (#12).
• Failed root sampling reports why: IK unreachable, in collision, or
  constraint violated.
• Integration test on the UR5e in MuJoCo (skipped without the menagerie),
  hypothesis property tests for the set laws, and headless smoke tests for
  the examples.
• MIT license, SPDX headers, issue and PR templates.

Changed

• Sampling from TSRs is reproducible under a seed. 1.0.0 used numpy's
  global random state for poses, so seeded plans were not repeatable.
• Goal and start bias apply to any sampleable set, including finite sets.
• With several constraint TSRs, projection converges each TSR before
  re-evaluating which is most violated. Single-TSR constraints are unchanged.
• Bias samples are collision-filtered by the validator rather than by
  IKSolver.solve_valid; only solve is required of a solver for sets.
• TSR-sampled goal configurations are collision-checked before becoming tree
  roots.
• The tsr package is imported only by pycbirrt.tsr_set and
  pycbirrt.legacy.

Deprecated

• CBiRRTConfig.tsr_tolerance. Passing it sets membership_tolerance and
  connection_tolerance together and emits a DeprecationWarning; reading
  it returns membership_tolerance.

Fixed

• Constraint projection with non-identity TSR frames. The closest
  in-bounds point was used as a world pose without composing T0_w and
  Tw_e, so projection failed whenever a constraint TSR had a translated or
  rotated frame (#28).
• Do not mark bounded joints as angular. The UR5e examples marked its
  ±2π joints angular, which let the planner join configurations a full turn
  apart and spin a joint through 360° on execution (#35). angular_joints
  is for genuinely unlimited joints only; the config comment now says so.
  If you copied the old example configuration, remove angular_joints.
• examples/planar_arm.py and examples/multi_config_demo.py crashed on a
  result field removed in March (#32).
• Collision checker state no longer goes stale between queries (#11).
• CI: the formatter version is pinned to the 0.15 series and a stale lock
  file that could never be honored was removed (#26).

Known limitations

• On joints with a range wider than 2π, TSRConfigurationSet offers only
  the IK solver's representative of each branch, not its in-limit 2π twins,
  so a short route through the twin can be planned as a long one (#36).
• TSR chains are not yet supported (#7).