Accurate transformation of pitch dimensions

[probberechts]

This PR is a complete rewrite of the "PitchDimensions" domain model. It implements the following main features.

1. It implements piecewise linear transformations in each dimension (see Stats Perform MA3 docs, appendix 7) to reduce the error when transforming between different coordinate systems.

transformer = DatasetTransformer(
    from_pitch_dimensions=OptaPitchDimensions(),
    to_pitch_dimensions=MetricPitchDimensions(
        x_dim=Dimension(0, 105),
        y_dim=Dimension(0, 68),
        pitch_length=105,
        pitch_width=68,
        standardized=False,
    ),
)

transformed_point = transformer.change_point_dimensions(
    Point(17, 78.9)   # the corner of the penalty area...
)
# ... remains the corner of the penalty area
assert transformed_point == Point(16.5, 54.16)

2. It makes it possible to compute distances between two points without having to transform the entire dataset to a metric coordinate system.

pitch = OptaPitchDimensions(pitch_length=105, pitch_width=68)
distance = pitch.distance_between(Point(0, 50), Point(11.5, 50))
assert distance == 11  # the distance from the goal line to penalty spot should be 11m

3. Fixes Different unit used for shot_event.result_coordinates.z dimension for Opta vs StatsBomb #293
4. Allows creating pitch dimensions with unknown boundaries. This will allow solving Incorrect Stats Perform coordinate system #299 .

BREAKING CHANGES:

This changes the interface of the PitchDimensions class.

[koenvo]

This seems like a bit of a magic number here. Can it be moved to some other place or to a 'constant' ?

[probberechts]

I've replaced it by a constant + optimized it a bit more to avoid unneeded conversions between dimensions.

[koenvo]

One minor comment. Other than that (and the merge conflict) happy to merge!