ENH: Simplify convert_math_to_symbolic

[bocklund]

Some changes to convert_math_to_symbolic following on to discussion in #4. I think the root cause is of the weirdness is that convert_math_to_symbolic is oddly polymorphic, where math_nodes can be strings or nodes. I tried to split it up a little bit so there's not a circular call graph where convert_math_to_symbolic(with interval nodes) -> convert_intervals_to_piecewise -> convert_math_to_symbolic(with a string node). I'm not sure if it makes sense without breaking the intent of how to handle intervals (and possible nesting of intervals, although I think that is currently broken).

If my understanding is correct, I think it should be that any incoming node to convert_math_to_symbolic will have:

1. Zero or more intervals, or
2. A text expression (where empty text ('') implies zero) that is additive to (1).

Is there a practical difference between how we treat the Interval text (''.join(interval_node.itertext()).replace('\n', '').replace(' ', '').strip()) vs. the Expr/Parameter text (''.join(node.xpath('./text()')).replace('\n', '').replace(' ', '').strip())? Maybe for nested intervals?

The current test suite is admittedly small, but I verified that the text in both cases is the same across the test suite (if you use xpath on interval_node). Except for cases where S.Zero != Float(S.Zero), the current test suite still passes after this change. I have a corresponding pycalphad patch lined up to coerce all numeric values in Gibbs energy expressions to symengine Floats in the DAT.