TODO

Known issue

Closed paths are not yet rendered correctly when the final (\pgfpathclose) segment is an edge case. There are multiple reasons for this:
1. invisibleline and closepath segments must be checked for the edge case, which is not implemented yet
2. invisibleline and closepath segments must be shortened on both ends, so their start and end angles must be set correctly. Furthermore, they must be shortened even though their previous/subsequent segment is formally invisible. Neither is implemented
3. If the above were implemented, there may still be issues left to solve

consider splitting tikzlibrarynfold.code.tex into a pgf and a TikZ library, as most of it does not need TikZ
look for potential issues with large paper sizes

angle too sharp detection: Is there a better way to detect if we move a point back too far?
- May be possible for lines but very hard for curves
do curvature checks and throw warnings at the start and end of offset segments

remove direct and indirect \pgfmathparse calls, mostly in the joins
- check if \pgf@nfold@roundjoin can be simplified before optimising
add \else in various places (likely better performance)
the A1A4 length is computed for every offset; this is something we could cache from the subdivision phase
faster, less precise veclen@ -> this is being used a lot
- if x > y, compute x*(1+(y/x)^2)^(1/2) approx x*(A+B*(y/x)^2); optimize A=1. and B=.5 over the unit circle
- could copy a lot of the existing veclen code; maybe skip the "scaling for precision" steps?
rendering: are quick commands an option? qmoveto, qlineto etc.?