Make matrix assembly and block-implicit time integration more helpful

It's too hard to debug matrix assembly. These things should help.

• print the structure of binary assembler expressions, if that only means printing the left and right structures
• check for Inf and NaN during assembly if in Debug and not on GPU
• add a stdout debug output to the MatrixExpression so we can attach that as a root expression to an integrator tree and get output easily on any assembler
• add an example of printing matrix output to debug dual time integration
• disable matrix debug output if threads are enabled