Create LazyFloat64 type for units

[MilesCranmer]

This PR creates the LazyFloat64 type for storing units. It makes units try to demote their numeric type, rather than promote everything to Float64:

julia> typeof(1u"m/s")
Quantity{Int64, Dimensions{DynamicQuantities.FixedRational{Int32, 25200}}}

julia> typeof(1.0u"m/s")
Quantity{Float64, Dimensions{DynamicQuantities.FixedRational{Int32, 25200}}}

julia> typeof(1f0u"m^2.5")
Quantity{Float32, Dimensions{DynamicQuantities.FixedRational{Int32, 25200}}}

julia> typeof(1u"fm")
ERROR: InexactError: Int64(1.0e-15)

@mcabbott could you review this?

[MilesCranmer]

This ensures that any unit is converted to a LazyFloat64 at the end. So even u"1.0" would become a LazyFloat64.

[github-actions]

Benchmark Results

	main	a1cfbbb...	t[main]/t[a1cfbbb...]
creation/Quantity(x)	3.2 ± 0.1 ns	2.8 ± 0.1 ns	1.14
creation/Quantity(x, length=y)	3.3 ± 0.1 ns	3.3 ± 0.1 ns	1
time_to_load	0.139 ± 0.0029 s	0.145 ± 0.003 s	0.961
with_numbers/*real	2.9 ± 0.1 ns	3.2 ± 0.1 ns	0.906
with_numbers/^int	21 ± 1.6 ns	20.7 ± 1.7 ns	1.01
with_numbers/^int * real	21.1 ± 1.7 ns	21.1 ± 1.3 ns	1
with_quantity/+y	5.9 ± 0.1 ns	5.7 ± 0.2 ns	1.04
with_quantity//y	3.3 ± 0.1 ns	3.7 ± 0 ns	0.892
with_self/dimension	1.6 ± 0.1 ns	1.6 ± 0.1 ns	1
with_self/inv	3.3 ± 0.1 ns	3.3 ± 0.1 ns	1
with_self/ustrip	1.6 ± 0.1 ns	1.6 ± 0.1 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).

[MilesCranmer]

Actually I think it's better if we only demote to abstract floats, but try to promote for everything else. Then we don't have to worry about integers.

[MilesCranmer]

Moved to #66