Add generic initializers for statically-dimensioned array types

[thatcomputerguy0101]

This takes advantage of const generics to provide general initializers for the statically-dimensioned array types. (2-6 dimensions, with the current implementations). This reduces the reliance on macros for array initialization, and enables other crates to operate more succinctly when working with an unknown-dimension static array type. I'm unsure how well this fits with the existence of the dynamic array index implementation, and the name of the new type alias IxD is open for discussion. No additional test cases have been implemented for this functionality yet.

[bluss]

Hm, new clippy lints should be fixed in a different PR, I think

[thatcomputerguy0101]

I originally decided on the name IxD to match the style of Ix0, Ix1, etc..., but with a generic D instead of a fixed value. It is also short, which helps in its usability as a convenience type (since it is an alias for Dim<[Ix, D]>). However, its name is close to IxDyn, and I have concerns about possible confusion between those two.

[thatcomputerguy0101]

With my most recent updates, I believe all static-related simplification that can be done within the current Rust feature set is implemented. This can be revisited in the future once rust-lang/rust#76560 lands.

[thatcomputerguy0101]

I added some basic tests for the new initializers. The other changes should all be covered by the existing tests since they were just implementation refactorings.

[thatcomputerguy0101]

I was having some problems with getting rust-analyzer in VS Code on my system to correctly see the new test file, but it seems to be picked up by command line clippy and cargo test.

[bluss]

I'm sorry about the delay here.

I now understand that this PR adds IxD (quibble: too similar to the existing name IxDyn).
IxD<2> has the same job as Ix2, only the first is const generic.

We can't accept this change at the moment, we don't want an aliasing dimensionality - two different ways to create a (statically known) two dimensional ndarray.

However, of course we need to use const generics better, and incremental changes is the way to get there, but I can't see us taking this path.

Something fruitful would be removing Ix2 and replacing with a better implementation, while preserving backwards compat with existing code as much as possible. (Introduce new type alias Ix2 and function Ix2?).

[thatcomputerguy0101]

Yeah, I had concerns about the name too. This is not adding a distinct type, only an additional alias, so Ix2 and IxD<2> are already the same type. This just relates all of the static array types under the IxD name (and adds the repeating factory). Technically, most of these changes could instead be implemented directly on Dim<[Ix; D]>, although that is less discoverable than giving it a dedicated name.

[thatcomputerguy0101]

I suppose underlying the exact methodology, I am seeking a way to access arbitrary diagonal elements of an unknown statically-dimensioned ndarray. I was doing this with a combination of IxD::repeating(i: Ix) and IxD::new(ix: [Ix; D]), but the same effect could also be achieved if indexes had a map function similar to the built-in static arrays. Would this be a change you are more open to accepting?

[thatcomputerguy0101]

Actually, I missed that basic arrays already implement IntoDimension, which might suit my needs.

[thatcomputerguy0101]

Seems like that does work; I think I tried that before, but was missing a trait constraint in my generalized code.