Refactor inverse_eltype calculations to use types.

[tpapp]

This is a continuation of #133 in a sense, using the approach to fix some other issues.

Incidental changes:

• whitespace and formatting
• clarify that inverse_eltype does (may) not do input checks
• fixed Inverse transformation for empty vectors fails #73, the corner case of the empty vector; tests added
• rename utility function introduced in Make inverse(::VectorTransfrom, x) return a vector of floats #133 and place in utilities.jl

[devmotion]

I'll review in more detail later today or tomorrow. Generally, it looks fine but this is a breaking change (inverse_eltype is part of the documented interface and with this PR inverse_eltype of TransformTuple etc call the definition with types which mostly likely downstream packages do not define currently and hence will throw an error).

[tpapp]

Yes, of course it is a breaking change. That's why I released 1.0.0 yesterday, instead of relying on "you can break everything in 0.x".

Among registered packages, the following are affected because they define a method (based on dependents list and code search):

• https://github.com/ptiede/HypercubeTransform.jl
• https://github.com/cscherrer/MultivariateMeasures.jl
• https://github.com/stjordanis/Pumas.jl
• https://github.com/ptiede/Comrade.jl
• https://github.com/JuliaMath/MeasureTheory.jl
• https://github.com/ptiede/VLBIImagePriors.jl

If we release 2.0.0, we might as well fix #66 since that is breaking too.

[tpapp]

I decided pull back the 1.0.0 release, release this as 0.9.0, and fix other breaking changes before 1.0.0.

@stjordanis, I am pinging you here since the above Pumas.jl repo does not accept issues.

[devmotion]

You don't have to worry about that repo, it's a completely outdated version of Pumas that was copied by the Github user when its source code was publicly available.