In the DCEL data structure, there are labels on the vertex records,
(half) edge records, and face records. These labels keep track
information such as which input geometry the record originates from and
whether it will appear in the output geometry.

The original approach to labelling was to use the same data structure
for all record types. The label had two booleans per operand.  The first
boolean determined whether the record would appear in the final output,
and the second boolean would show whether the first boolean was
populated or not. This was essentially tri-state logic. The trouble with
this approach was that the DCEL algorithm doesn't know whether each
record should appear in the output geometry right up until the last
stage. The labels are used in the earlier stages of the algorithm to
record other facts, such as whether or not the records appear in the
input geometry. The semantics of the fields subtly change over the
course of the algorithm, making the code hard to understand and
maintain.

The new approach (implemented by this PR) changes how label management
is performed. Instead of having a single labels data structure, the
fields are slightly different for vertex, edge, and face records. This
is to account for the differences in information available when
processing vertexes vs. edges vs. faces. The number of fields is also
expanded, so that concepts like "is this record explicitly present in
the input geometry's control points" are separate from concepts like "is
this record present in the input geometry but not explicitly in its
control points". The populated fields are also removed, and tracked
separately in each algorithm that modifies the labels.