Clean-up modcc output

[thorstenhater]

1. add a series of algebraic simplifications to NMODL output

• -(-a) == a contraction
• -(a - b) == b - a
• ...

2. add detection and emission of fused-mul-add operations
3. handle all SIMD constraint from Arbor in NMODL
4. replace more << output with fmt for cleaner reading

These can yield faster code, but doesn't in my local benchmarking, likely because the
C++ compiler is smart enough to spot these. However, it makes the output cleaner and
might help the odd compiler.