Create QuantityArray <: AbstractArray

[MilesCranmer]

This creates a type:

QuantityArray{
    T,
    N,
    D<:AbstractDimensions,
    Q<:AbstractQuantity{T,D},
    V<:AbstractVector{T,N}
} <: AbstractArray{Q,N}

that stores a single set of physical dimensions for the entire array.

For example:

A = QuantityArray(randn(32), u"km/s”)

it implements a custom broadcasting interface so as to create a new QuantityArray with the correct output dimensions. This part is a bit involved as it needs to materialize the first element. Thanks to tictaccat on Discourse for help.

Everything is type stable.

For safety purposes, the compiler is in charge of figuring out when the dimension calculation can be skipped for each element or not. It should be able to if it inlines things correctly, which happens for simple broadcasting. For more complex broadcasting it might not be able to inline things.

@ChrisRackauckas do you think you or someone else in @SciML could review this to see if it fits your requirements?

[github-actions]

Benchmark Results

	main	c1bdf62...	t[main]/t[c1bdf62...]
Quantity/creation/Quantity(x)	3.2 ± 0.1 ns	3.2 ± 0.1 ns	1
Quantity/creation/Quantity(x, length=y)	3.3 ± 0.1 ns	3.7 ± 0.1 ns	0.892
Quantity/with_numbers/*real	2.9 ± 0.1 ns	3.2 ± 0.1 ns	0.906
Quantity/with_numbers/^int	21 ± 1.7 ns	10.1 ± 2.5 ns	2.08
Quantity/with_numbers/^int * real	21.1 ± 1.5 ns	10.5 ± 2.4 ns	2.01
Quantity/with_quantity/+y	6.7 ± 0.3 ns	6 ± 0.3 ns	1.12
Quantity/with_quantity//y	3.3 ± 0.1 ns	3.7 ± 0.1 ns	0.892
Quantity/with_self/dimension	1.6 ± 0.1 ns	1.6 ± 0.1 ns	1
Quantity/with_self/inv	3.3 ± 0.1 ns	3.7 ± 0.1 ns	0.892
Quantity/with_self/ustrip	1.6 ± 0.1 ns	1.6 ± 0.1 ns	1
time_to_load	0.136 ± 0.0012 s	0.165 ± 0.0014 s	0.822

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

[MilesCranmer]

Seems like a bunch of the array tests pass on 1.10 but don't on 1.9... Maybe something in the broadcasting interface changed.

[MilesCranmer]

Woo! 100% testing coverage again!

I'll do another code review and then we can merge.

[MilesCranmer]

My one comment is that these should probably be turned into promote_rule at some point. But that's for another PR.

Also I realize this specific and unusual promotion requirement might make things a bit harder to move FixedRational into Ratios.jl (which I still think is a good idea, cc @oscardssmith)... But we can just figure that out later.

[MilesCranmer]

Thanks again for your help @gaurav-arya!!