Fix coordinates-to-coordinates score() passing kwargs __call__ rejects

[jkosinski]

Problem

optimize_match on any coordinates-to-coordinates score (Chamfer, NormalVectorScore) raises:

TypeError: __call__() got an unexpected keyword argument 'transformed_coordinates'

_MatchCoordinatesToCoordinates.score() calls

return self(
    transformed_coordinates=self.template_coordinates_rotated,
    transformed_coordinates_mask=self.template_mask_coordinates_rotated,
)

but the concrete __call__ methods (Chamfer, NormalVectorScore) take no arguments — they
read self.template_coordinates_rotated, which _rigid_transform(..., out=self.template_coordinates_rotated)
has already written. So the kwargs are spurious and break the call.

Fix

Make the coordinates-to-coordinates score() call self() with no kwargs, matching the
coordinates-to-density base (_MatchCoordinatesToDensity.score, which does return self()).

Effect

Chamfer now optimises through optimize_match. NormalVectorScore still needs equal-size
pre-paired point clouds (its __call__ multiplies template_coordinates_rotated * target_coordinates
elementwise) — a separate issue, noted but not addressed here.

[jkosinski]

Out of scope for this PR, noted for later: NormalVectorScore.call does not multiply its result by self.score_sign (unlike Chamfer and the other scores), so its sign convention is inconsistent. Can be fixed in a follow-up.