Implemented strides for AbstractArray subtypes

[raghav9-97]

Fixes #29705

[mbauman]

Thanks for the contribution! We should probably make the documentation of strides a bit more clear — it doesn't give you the multipliers you need to compute a linear index, but rather it defines how to find a given location in memory as an offset from a base pointer.

This means that, for example strides(view(rand(10), 1:2:10)') should be (1, 2) and not (1,1). It also means that we cannot support strides on Diagonal as its memory layout does not conform to the StridedArray interface.

See https://docs.julialang.org/en/v1/manual/interfaces/#man-interface-strided-arrays-1 for a bit more detail — but probably still not quite enough.

[mbauman]

Define strides(::AdjOrTrans) in terms of the strides of its parent, and remove the strides definition for Diagonal

[raghav9-97]

@mbauman If i am correctly understanding strides

A = rand(3,5)

strides(A) should generate (1,3) and strides(A') will generate (1,5) or will it generate (3,1) just as similar to PermuteDimsArray(A, (2,1))?

[mbauman]

It's even simpler than that — you just want to reverse the strides of the parent. Oh my, now that I write that out, it's sounding very familiar to me…

Yup, shucks. #29135. I have a really bad habit of submitting PRs and then forgetting all about them.

[raghav9-97]

@mbauman Haha!! You've already implemented it way before than me?Should I continue working on this?Also should I split the function for StridedMatrix and StridedVector as you did?

[ViralBShah]

@mbauman Should we merge your old PR, or should @raghav9-97 prepare this one to be merged?