Error on overflow in dimensions variable

[MilesCranmer]

Ratios.jl is faster, but it makes rational numbers much more susceptible to overflows. This switches from Int to SafeInt, which will result in overflow errors when they occur.

I think this should have been done when Ratios.jl was first added to this repo. It's just the safer thing to do; otherwise you could get awful silent bugs.

@oscardssmith what do you think? This makes it a bit easier to use SafeInt8 as well. As before I think the user should just be able to parametrize their Dimensions type, with SimpleRatio{SafeInt} being a sensible default.

[github-actions]

Benchmark Results

	main	8fe7dc7...	t[main]/t[8fe7dc7...]
creation/Quantity(x)	4.1 ± 0.1 ns	4.1 ± 0.3 ns	1
creation/Quantity(x, length=y)	5.3 ± 0.1 ns	5.2 ± 0.4 ns	1.02
time_to_load	0.0857 ± 0.0012 s	0.164 ± 0.00091 s	0.524
with_numbers/*real	4.1 ± 0.001 ns	4.1 ± 0.3 ns	1
with_numbers/^int	20.5 ± 2.6 ns	0.0345 ± 0.0038 μs	0.594
with_numbers/^int * real	0.0398 ± 0.0046 μs	0.0341 ± 0.0042 μs	1.17
with_quantity//y	19.3 ± 0.5 ns	24.9 ± 0.8 ns	0.775
with_quantity/^y	1.4 ± 0.29 μs	1.55 ± 0.31 μs	0.898
with_self/dimension	2.1 ± 0.1 ns	2 ± 0.2 ns	1.05
with_self/inv	2.8 ± 0.2 ns	22.6 ± 0.1 ns	0.124
with_self/ustrip	2 ± 0.1 ns	2 ± 0.2 ns	1

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).