WIP: Towards a more "Julian" package

[timholy]

This is a proposed redesign that may be a bit more natural to work
with from the Julia perspective. The principal differences are:

• Support eltypes of the form SVector{T,N}: this makes this wrapper
  behave more like OpenCV itself, which (unlike Python) does not use
  an array dimension to encode the number of color channels.

• Support N0f8 and N0f16, JuliaImages preferred interpretation of
  8- and 16-bit unsigned intensity data. (See FixedPointNumbers and
  https://juliaimages.org/latest/tutorials/arrays_colors/#fixedpoint;
  interestingly, OpenCV itself suffers from the exact same problem, see
  https://stackoverflow.com/questions/14539498/change-type-of-mat-object-from-cv-32f-to-cv-8u
  for an example of a user who got bit by the "divide by 255" rule.)

• Work towards support for multidimensional arrays (see
  Multidimensional images JuliaImages/OpenCV.jl#4)

• Validate arguments more carefully and improve correctness in a few
  key places

• Improve inferrability (avoid abstract types JuliaImages/OpenCV.jl#1)

This has not been tested (I don't know how, see JuliaImages/OpenCV.jl#3), so don't merge this unquestioningly.

[timholy]

CC @johnnychen94

[timholy]

Note that I didn't put colorspace into the Mat type: that's because OpenCV itself seemingly doesn't tell you how to interpret the values stored in the array. (IMO, this choice is troubling: since when has there been a recent computer language that says "here's 64 bits, good luck guessing whether it's a Float64 or an Int64. But now go ahead and do some math on it despite the fact that those two have totally different rules for the bit-level representation of their math.")

[archit120]

Hi Tim,

I should probably have archived this repository and will do that rn to avoid confusion. The changes that you made should go into https://github.com/opencv/opencv_contrib/tree/master/modules/julia/gen/jl_cxx_files instead