Currently, all decompositions can be called on integer matrices but usually generate bogus results, since they use floating point operations (division/sqrt) which are silently converted to integer results.
We should forbid such decompositions at compile-time.
Decompositions which only depend on division (LU/LDLt) should be allowed for rational-number implementations, however.
At least, we should document the requirements on the Scalar types and explicitly list the common supported types.
*** This bug has been marked as a duplicate of bug 949 ***