Support for VPA, gpuArray, and other special data types

[Steven-Roberts]

For most problems (especially vanderpol) it would be nice to support arbitrary data types. Here are some of the issues and things to consider

• If a parameter is of a certain type, should Y0, TimeSpan, and other parameters be cast to the same data type? Sometimes this cannot be deduced or there are multiple parameters with different types.
• cast is the most reasonable function to do this but functionality is limited in Octave. Critically, cast(x, 'sym') does not work.
• Numbers with more than ~16 digits have to be broken up before casting.
• Octave does not support the like name value pair used in functions like zeros and cast.
• When we create differentiation matrices, the type needs to be consistent.
• Initial condition checks in solveExactly may be fragile with arbitrary data types.