DoubleInequalityTheory uses BigDecimal type

[mtf90]

Contrary to its name, DoubleInequalityTheory is a TypedTheory<BigDecimal>. This causes issues when you want to set up type-safe theories in situations such as LearnMixedIOTest where the RA has Double-typed registers.

Currently, this doesn't cause any issues at runtime because JConstraint's Variables store the values in String representation and parse them via the specified BuiltinType once it is accessed (and BigDecimal can parse a Double-represented value). However, once the theory generates fresh values (BigDecimals), storing them in the respective RA may cause problems because both data types have different precision (and BigDecimal and Double do not have an inheritance relation).

I think there are two obvious solutions to this problem:

• Either switch the RAs to BigDecimal-based registers (requires some re-factoring of the XML models and the parser),
• or switch the DoubleInequalityTheory to actually work on BuiltinTypes.DoubleType. This could also be interesting from a performance point of view since operations on doubles should be faster than on BigDecimal.

I don't have a preference for any of the solutions, but we should do something about this. The current implicit conversion is primed for some hard-to-detect issues in the future.