Matrix subset according to type of input

[dvd101x]

Hi Jos,

This addresses #2344

It changes the behavior of index to accommodate for scalar indices and the size of the output depends on the input.

I'm still doing some tests and reviewing what documentation needs to change, but it all seems to work according to:
#2344 (comment)

A = [1,2,3; 4,5,6]
A[2,3]  --> 6
A[[1,2],[2,3]] --> [2,3; 5,6]
A[[2],[2,3]] --> [[5, 6]]
A[2,[2,3]] --> [5,6]
A[[1,2],[2]] --> [2; 5]  // which is the same as [[2], [5]]
A[[1,2],2] --> [2,5]
A[[2], [2]] --> [[5]]
A[2, [2]] --> [5]
A[[2], 2] --> [5]
A[[],[]] --> []
A[1,[]] --> []
A[[],1] --> []

[dvd101x]

I think it's mostly ready for review.

Just wanted to review the following,

The elements of an index are called ranges like index(range1, range2) but the ranges are not always Range. Sometimes ranges are number, string, bigNumber, Array, Matrix, Array of Booleans, Matrix of Booleans or Range.

So I was thinking if in the documentation and maybe in the code to find a more general name for the elements of an index like index(indexElement1, indexElement2).

As a reference in numpy it's more like you slice with indices and there is no index class. So you slice A[index1, index2]

https://mathjs.org/docs/reference/functions/index.html

[josdejong]

The code looks really neat, thanks David!

Some thoughts:

1. This is a breaking change, therefore I've changed this PR to merge into the v15 branch, and listed it at #3453.

2. So I was thinking if in the documentation and maybe in the code to find a more general name for the elements of an index like index(indexElement1, indexElement2).

   Maybe we can call the arguments index(dim1, dim2, ...)?

3. Does this PR have impact on the performance of subset?

4. I'm a bit concerned about how hugely breaking this change is. Existing code will probably still work but now return differently dimensioned matrices, which can give subtle and hard to debug errors. Can we somehow help people migrate, or detect "old" usage and warn the user? Or make it backward compatible with a config flag?

[dvd101x]

Thanks for your review Jos,

1. Ok.
2. Good idea! that aligns better with the rest of the code base.
3. Makes a few of the tests slightly faster within the margin of error, but the main speed benefit might be the Matrix creation during the creation of Index, which is outside the scope of the subset benchmark. I will add some tests including the Index creation.
4. I agree it's a big change. I'm not sure how to proceed. I'll give it some thought.