Improve test coverage

[ffreyer]

This pr aims to increase test coverage and fix problems I hit along the way. WIP for now.

Questions:

• Should Rect() default to Rect(NaN..., 0...) instead of Rect(Inf..., -Inf...)? Otherwise, how should utility functions treat infinite Rects? E.g. should volume return NaN? minimum and maximum? Should we consider -Inf widths as invalid?
• Does it make sense for mat[Rect2i(1, 1, 2, 2)] to index [2:3, 2:3], i.e. be zero based?

(Significant) (Code) Changes:

• Rect:
  • refactored constructors to generalize them and make them depend more on each other. (This should only add more constructors)
  • generalized euclidean_distance, update to accept VecTypes
• added strictly_equal_face_vertices() for comparing the result of split_mesh with pre-merge meshes. This considers two meshes to be equal if a.vertex_data[a.face] == b.vertex_data[b.face] holds for every face and every vertex attribute.
• removed merge_vertex_indices(faces...[, index_counter]) method which I added as an alternative to merge_vertex_indices(tuple_of_face_arrays[, index_counter]) but never used. This function is internal with the (more) public interface being expand_faceviews(), so this should not be breaking. (It's also not used in MeshIO or Makie)
• breaking Removed Rect{T}(obj) constructor for boundignboxes as it clashes with the Rect{N, T} definition of the type. The implementations now use Rect{N, T}(obj) with other methods funneling into that one. This enables Rect{N}(obj) and RectT{T}(obj) analogous to other Rect constructors.
• maybe breaking reworked self_intersections() to allow a segment to intersect with multiple other segments. For this, the format of the returned indices changed from Vector{Int} to Vecto{Tuple{Int, Int}}. This is now an internal change. Also did some low hanging performance optimizations (skip redundant half of the loop)

Test TODO;

• Rect (file: 57% -> 96%, total: 64% -> 71%)
  • constructors, convert, centered, split, bbox-like constructors
  • euclidean distances
  • Rect as matrix index
  • Matrix - Rect multiplications
• bits and pieces in basic_types.jl (file: 63% -> 83%, total: 74%)
• most of triangulation.jl (file: 26% -> 90%, total: 80%)
• meshes.jl (file: 68% -> 85%, total: 82%)
  • split_mesh
  • duplicate face removal
  • map_coordinates
• boundingboxes.jl (file: 48% -> 88%, total: 83%)
• lines.jl (file: 34% -> 98%, total: 86%)
• spheres.jl (file: 63% -> 100%, total: 86%)
• OffsetIntegers (file: 61% -> 55% ???) and FixedArrays (unchanged) (total: 86%)

[asinghvi17]

Do we actually have codecov enabled on this repo? Should we do that, if the coverage is now worth showing off? :D

[ffreyer]

We do, it just doesn't run in prs

[ffreyer]

Well it does, it just errors

[ffreyer]

I don't think there is much of a point to making a user facing depwarn here, because on master Rect{T}(geom) only works if T matches the input eltype. I therefore doubt anyone relied on it. It also throws an ArgumentError lower down the line (and would create a stackoverflow with this dep warning.)

For devs I think it makes more sense, because I think it's fairly likely someone would copy the syntax from GeometryBasics. And without a warning, their implementation would silently be ignored.

[ffreyer]

nvm, doesn't work. I thought for a second that N <: Val was working for catching literals but I don't think it actually does. Without type specifity on Rect{N}(...) this will always call the applicable branch and loop.

[ffreyer]

Fixes #155 (if it's not already fixed on master)

[ffreyer]

Fixes #207