Fix multiple linear/nonlinear solves

[landinjm]

The change variables were hard coded so that only the first TIME_INDEPENDENT equation would have its change variable (for Newton-Picard iteration). This prevented the addition of multiple TIME_INDEPENDENT (#166) equations. To fix this, I switched the FEEvaluation vectors to unordered maps instead.

This way, we can assign the FEEvaluation to the global variable index rather than the index of variable type (e.g. phi might be the 4th explicit variable and the 5th global variable. In the prior scheme we would index based on the 4th explicit). Doing so also reduced the convolutedness of loadVariableAttributes.cc and variableContainer.cc.

Some minor optimizations were also made in regard to dereferencing over and over. Before we had this

scalar_vars_map[var_index]->reinit(cell);
scalar_vars_map[var_index]->read_dof_values(*src[i]);
scalar_vars_map[var_index]->evaluate(varInfoList[i].evaluation_flags);

Instead, I stored the pointer in a local variable.

auto *scalar_FEEval_ptr = scalar_vars_map[var_index].get();
scalar_FEEval_ptr->reinit(cell);
scalar_FEEval_ptr->read_dof_values(*src[i]);
scalar_FEEval_ptr->evaluate(var_info.evaluation_flags);

Since these are called for every cell and every step, the performance improvement should be rather significant.