Optimize SpecModel predictions

[bd-j]

There are a number of potential inefficiencies in SpecModel.predict(), including multiple calls to functions that might not be cheap, reading emline_info from disk every call, and linear algebra operations on sparse matrices.

Also, add a public method to return emission line luminosities.

[bd-j]

Some of this was fixed by #219 (e.g. the emline_info caching) but there are still some linear algebra operations that might be made more robust or faster.