Add correspondence-quality command with Python API (#2612)

[akenmorris]

Closes #2612.

Adds a shapeworks correspondence-quality subcommand and matching Python API.

For each subject, reconstructs the surface from its local particles via a biharmonic mesh warp from the cohort medoid template (matching ParticleShapeStatistics::compute_median_shape, so it's the same subject Studio picks as median), and measures point-to-cell distance to the groomed mesh. Distances are also normalized by each subject's bounding-box diagonal for scale-invariant "good/bad" judgement across anatomies. R

CLI

shapeworks correspondence-quality --name <project.swproj|xlsx>
    [--output <csv>] [--output_meshes <dir>]
    [--method point-to-cell|point-to-point] [--worst N]

• Handles both mesh and distance-transform groomed inputs (DT → mesh via Image::toMesh(0.0)).
• Handles multi-domain projects (single global L1-medoid template across all domains, mirroring Studio).
• Aggregates (mean/median/p95/max) exclude the template's own near-zero row so small cohorts aren't skewed by it.
• Optional --output_meshes writes each reconstruction as .vtk with an embedded per-vertex distance field for inspection in Studio/ParaView.
• CSV columns: subject, domain, is_template, mean_dist, max_dist, bbox_diag, norm_mean, norm_max.

Python binding (sw.CorrespondenceEvaluation)

project = sw.Project(); project.load("model.swproj")
report = sw.CorrespondenceEvaluation.evaluate(project)
print(report.template_subject, report.agg_norm.mean)
for r in report.rows:
    print(r.subject, r.norm_mean, r.is_template)