Implement PyDenseArray

[JamesWrigley]

This is a minimal array type to allow dispatching to things like BLAS functions that require DenseArray, inspired by the idea proposed in #319 (comment). Example performance improvement:

julia> using PythonCall, BenchmarkTools, LinearAlgebra

julia> @py import numpy as np

julia> x_py = np.random.rand(1000, 1000);
                                                                                                                                                                                                                                               
julia> x1 = pyconvert(PyArray, x_py.T);
                
julia> x2 = pyconvert(PyDenseArray, x_py);

julia> @benchmark mul!(out, x1, x1) setup=(out=zeros(1000, 1000))
BenchmarkTools.Trial: 20 samples with 1 evaluation per sample.
 Range (min … max):  253.484 ms … 264.718 ms  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     259.075 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   258.614 ms ±   2.880 ms  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▁    ▁ █     ▁      █   ▁ ▁  ▁ ▁ █ █   ▁▁   ▁  ▁            ▁  
  █▁▁▁▁█▁█▁▁▁▁▁█▁▁▁▁▁▁█▁▁▁█▁█▁▁█▁█▁█▁█▁▁▁██▁▁▁█▁▁█▁▁▁▁▁▁▁▁▁▁▁▁█ ▁
  253 ms           Histogram: frequency by time          265 ms <

 Memory estimate: 0 bytes, allocs estimate: 0.

julia> @benchmark mul!(out, x2, x2) setup=(out=zeros(1000, 1000))
BenchmarkTools.Trial: 429 samples with 1 evaluation per sample.
 Range (min … max):  9.034 ms …  13.821 ms  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     9.213 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   9.270 ms ± 339.861 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

    ▅▁  ▁▄▃▃▅▄▆█▂                                              
  ▄▄██▇▇█████████▃▅▃▃▃▂▄▁▄▆▆▅▄▄█▇▂▄▁▂▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▂▁▁▁▁▁▁▁▂ ▄
  9.03 ms         Histogram: frequency by time        9.97 ms <

 Memory estimate: 0 bytes, allocs estimate: 0.

I think this makes more sense than a DenseView type in a separate package because this way in v1 we can make it the default instead of PyArray when converting contiguous arrays. Which would be very nice for performance :) It ended up being a fairly small amount of code since it's such a restricted type, so hopefully it's not too much of a maintenance burden.

Written with help from Claude 🤖