logabsdet test fix

[mzgubic]

Closes JuliaDiff/ChainRulesCore.jl#266

There were two issues:

1. missing ;kwargs causing the wrong method to dispatch
2. finite differencing returns a Tuple for functions which output a Tuple

[oxinabox]

Needs tests.
We can make a MWE of a rule for x->(x, x) where test_frule currently errors/fails

[mzgubic]

right, good to keep the test inside ChainRulesTestUtils and not rely on integration to flag this.

I needed to do x->(x, 1.0) because it is important for the backing type of the composite to be different from the differential so that it doesn't _can_pass_early

[oxinabox]

using #TODO causes it to get picked up by various automated tooling

Suggested change

	# remove after https://github.com/JuliaDiff/FiniteDifferences.jl/issues/97
	#TODO remove after https://github.com/JuliaDiff/FiniteDifferences.jl/issues/97

[oxinabox]

also maybe a comment about this fixing up FiniteDiff sometimes returning Tuples etc rathter than valid diff types as part of that TODO.
to explai what this function does?

[oxinabox]

Since they are not expored

Suggested change

	to_composite(x::Tuple) = Composite{typeof(x)}(x...)
	to_composite(x::NamedTuple) = Composite{typeof(x)}(;x...)
	to_composite(x) = x
	_to_composite(x::Tuple) = Composite{typeof(x)}(x...)
	_to_composite(x::NamedTuple) = Composite{typeof(x)}(;x...)
	_to_composite(x) = x

Might want to say:

Suggested change

	to_composite(x::Tuple) = Composite{typeof(x)}(x...)
	to_composite(x::NamedTuple) = Composite{typeof(x)}(;x...)
	to_composite(x) = x
	_maybe_as_composite(x::Tuple) = Composite{typeof(x)}(x...)
	_maybe_as_composite(x::NamedTuple) = Composite{typeof(x)}(;x...)
	_maybe_as_composite(x) = x

or something?

[oxinabox]

or maybe including fix in the name?

[mzgubic]

all good suggestions, so added them all