Refactor connect edges

[bluenote10]

This PR follows the ideas of w8r/martinez#113, combined with a few more fixes, and translated into a more Rust friendly structure. Basically the JS implementation was missing a key step of the Martinez algorithm in the second stage -- the implementation of the 4 possible cases for determining parent contours (Fig. 4 in the paper):

The idea is that for each start point of a contour we look at the lower ("previous in result") contour and check:

• if there is no contour below => current contour must be an exterior contour (case 1).
• if there is a lower contour, but it is an inside-outside transition => current contour must also be an exterior contour (case 3).
• if there is a lower contour, and it is an outside-inside transition => in this case, the current contour is a hole, and depending on whether the lower contour is also a hole or an exterior contour we can establish the parent/hole relationship (cases 2 & 4).

I also tried to document the ideas with code comments as best I could.

This PR fixes / affects:

• Union does not create polygon with holes #7
• Misattributed hole with intersection w8r/martinez#96
• intersection returning invalid polygon w8r/martinez#93
• Union show hole as multipolygon w8r/martinez#71
• Incorrect inner ring added to diff result. w8r/martinez#69
• outer ring added to wrong polygon in diff() result. w8r/martinez#68
• Invalid polygon in expected result of test 'disjoint union testing' w8r/martinez#61
• The 'fatal3' test case, which now has proper holes instead of all exterior rings.
• A few more new test cases involving polygon nesting that were tricky to get right.

Here is a visualization of all new / modified test cases:

test_cases.pdf

A few minor changes:

• The biggest debugging struggle for me was caused by the inability to display/debug print the Float trait. Omitting the Debug/Display traits from Float doesn't do anyone a favor who gets to debug Float based code. Hope it is fine to extend the trait, because Float will be almost exclusively instantiated by f32 or f64 anyhow.
• I've added the Python plotting script I use for visualizing the test cases in a scripts subfolder. It is ~200 LOC and I thought it might be useful for anyone testing the algrithm. I have also documented how the test scheme works in a short readme.

[untoldwind]

I admit that I do not entirely get the changes to the algorithm, but that's what test-cases are for ;)

I did a minor code-cleanup in master, mostly involving formatting and clippy.
On a side-note:
It's always a good idea to run
cargo fmt
and
cargo clippy
regularly.
Especially the latter gives a lot of insights

[agersant]

Thanks a lot for working on this @bluenote10! These bugs were the roadblock for me to consider replacing my usage of gpc with a Rust solution.