Add Hungarian and Sinkhorn point cloud matching methods

[Marei33]

Summary

Extends point cloud matching beyond the ILP solver by adding two alternative
matching algorithms, selectable via config. Also generalizes the ILP config
block into a method-agnostic matching block and cleans up the Snakemake rules.

Changes

New matching methods (matchmaker/utils/matching.py)

• hungarian_matching — optimal one-to-one matching via
  scipy.optimize.linear_sum_assignment on the dense euclidean distance matrix.
  Pairs beyond max_dist are dropped, yielding a partial matching like the ILP matcher.
• sinkhorn_matching — soft assignment via pygmtools.linear_solvers.sinkhorn
  on a distance-derived similarity matrix, discretized to a one-to-one matching with
  the Hungarian algorithm. Configurable via tau (entropy regularization) and max_iter.
• Renamed utils/ilp_matching.py → utils/matching.py (ILP matching retained unchanged).

CLI & pipeline (match_pointclouds.py, registration.smk)

• match_points / main gain --method {ilp,hungarian,sinkhorn}, --tau,
  and --sinkhorn_max_iter options (defaults preserve current ILP behaviour).
• Renamed Snakemake rule ilp_matching → matching; it now passes the method
  and Sinkhorn parameters through to the CLI.
• Removed the hard-coded configfile: line from registration.smk (config is now
  supplied on the command line).
• Dropped the conda: "matchmaker_env" directive from all rules.

Config

• Replaced the ILP: config block with a matching: block that adds a method
  key (ilp | hungarian | sinkhorn) across the example configs.

Dependencies

• Added pygmtools to environment.yml (required for Sinkhorn matching).