Use is_forwarding_wrapper trait instead of parent_type(T) <: T

[Tokazama]

Using the heuristic parent_type(T) <: T sometimes gives incorrect results and may hide holes in the interface (as I found when making these changes).

Some problems I found:

• is_increasing(stride_rank(A)) didn't account for A having zero dimensions
• is_dense(A) didn't know how to handle when dense_dims(A) returned nothing.
• contiguous_axis(::Type{<:ReshapedArray}) would sometimes return nothing (meaning the contiguous axis is unknown) instead of StaticInt(-1) when we actually knew there couldn't be a contiguous axis because it didn't exist in the parent array.
• (known)_dimnames(A) was often returning incorrect results for ReinterpretArray and ReshapedArray.

[codecov]

Codecov Report

Merging #296 (fe9307b) into master (a61ed56) will decrease coverage by 1.05%.
The diff coverage is 73.26%.

❗ Current head fe9307b differs from pull request most recent head f9410a1. Consider uploading reports for the commit f9410a1 to get more accurate results

@@            Coverage Diff             @@
##           master     #296      +/-   ##
==========================================
- Coverage   91.07%   90.02%   -1.06%     
==========================================
  Files           9        9              
  Lines        1345     1393      +48     
==========================================
+ Hits         1225     1254      +29     
- Misses        120      139      +19     

Impacted Files	Coverage Δ
src/axes.jl	92.56% <50.00%> (+0.25%)	⬆️
src/indexing.jl	87.86% <50.00%> (+0.48%)	⬆️
src/dimensions.jl	85.55% <56.25%> (-11.61%)	⬇️
src/stridelayout.jl	90.24% <93.10%> (+0.21%)	⬆️
src/ArrayInterface.jl	85.95% <100.00%> (-0.12%)	⬇️
src/size.jl	92.13% <100.00%> (+1.77%)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a61ed56...f9410a1. Read the comment docs.