Add generic test cases + switch to robust predicates

[bluenote10]

This PR mirrors:

• Use robust-predicates in signed area w8r/martinez#111 which has added robust predicates for the point-on-line check (signed_area). I'm not fully sure whether using robust predicates in this case is the best solution (because of this observation), but for now I would say it makes sense to stay close to the JS implementation.
• Added generic test case handling w8r/martinez#117 which introduces a GeoJSON based test suite that replaces all the manually coded edge cases and allows for easier visualization of the test cases. This step has revealed that some of the test cases were actually a bit broken.

A test case file is a GeoJSON file that contains both the input and the expected output, as explained here. Tests only succeed if the implementation produces exactly the specified values. The output data can be regenerated by running with REGEN=true cargo test (identically to REGEN=true npm run test on JS side). The big benefit is that no code has to be written for testing edge cases, and we can get a simple overview over what we test against (*):

test_cases.pdf

Differences to the JS PR:

• I had to disable two test cases (collapsed_edges_removed.geojson and disjoint_union_nesting.geojson) for Rust, but basically because these tests are quite broken anyway (the expected result on JS side isn't even valid GeoJSON), see the PDF attached to #117. These cases have to be revisited later anyway.
• Two of the test cases needed a tiny numeric adjustment after copying the test case from JS. I checked the differences and they are all caused by numerical fluctuations -- I guess that has to be expected, e.g. there could be a tiny difference in operator associativity between the implementations.

Notes on the Rust implementation: I haven't found a way to make the dynamically discovered test cases show up as separate tests on cargo level (I think the available #[test]'s must be statically known). This is a bit unfortunate, because now the first assert_eq failure will break the loop over test cases. We could use a boolean any_test_failed, but then we won't get the nice diffing of pretty_assertions::assert_eq.

(*) If desired I can commit the plotting script as well. It is a tiny Python script with usage plot_test_cases.py <geojson1> <geojson2> ....

[bluenote10]

Function name isn't fully appropriate anymore => rename to run_generic_test_case or so.

[untoldwind]

run_generic_test_case sounds about right

[untoldwind]

Apart from that minor refactoring looks good to me.

Would you like to fix that or should I just merge and do the refactoring myself?

[untoldwind]

Actually I'd write this part more like this:

#[inline]
pub fn coordinate_to_robust<F>(p : Coordinate<F>) -> Coord 
where
    F: Float,
{
    Coord{x: p.x.to_f64().unwrap(), y: p.y.to_f64().unwrap()}
}

#[inline]
pub fn signed_area<F>(p0: Coordinate<F>, p1: Coordinate<F>, p2: Coordinate<F>) -> F
where
    F: Float,
{
    let res = orient2d(
        coordinate_to_robust(p0),
        coordinate_to_robust(p1),
        coordinate_to_robust(p2),
    );
    if res > 0f64 {
        F::one()
    } else if res < 0f64 {
        -F::one()
    } else {
        F::zero()
    }
}

... but that is more like a personal preference

[bluenote10]

Looks good to me, I have quickly pushed all the changes...

[untoldwind]

run_generic_test_case sounds about right