Verifying Ginkgo with custom value types

[gflegar]

Currently we are only compiling Ginkgo for default (float, double, std::complex<> of those) value types when checking the correctness of our implementation. (We're also testing only for double, see #32 for discussions on that).

This means that the interface that ValueType needs to have may change after a PR without anyone noticing (e.g. someone used a part of the interface of default types, which user-defined types may not have to implement).

I propose we create a dummy ValueType implementation (we would create a concept instead if TS concepts were a part of C++11), that could just be a double wrapped in a class, and have the interface that Ginkgo's ValueType has to satisfy defined. This gives us two benefits:

• we can try to compile with this type whenever we test Ginkgo and get a compiler error whenever we inadvertently assume something new about ValueType's interface
• users can look into this type, and see what their custom value type needs to have defined in order to work with Ginkgo

[gflegar]

Whoever is going to implement this, they might find the description in #49 useful. It give an example of compiling Ginkgo with user-defined value types (in this case FloatX).