Allow NISTMultilineTRL to work with non-exact floats

[eendebakpt]

The NISTMultilineTRL checks whether variables are scalar with the construction

try:
   v = x[0]
except TypeError:
    # treat x as scalar
    ...

This works for float, since v = 1.0; v[0] results in a TypeError. But for v = npy.float64(1.0); v[0] an IndexError is raised. This PR allows NISTMultilineTRL to work with numpy scalars by checking arguments with numpy.isscalar and converting if required. The PR also improves performance of s2t by factoring out a common expression.

[Ttl]

I would prefer making them arrays in __init__ so that try except is not necessary. npy.isscalar could be useful for it as it handles both numpy types and ordinary python floats.

Just adding a comment is not good enough fix for #889. I'll submit a PR for it later.

[Ttl]

Looks mostly fine, just few comments.

[Ttl]

Could this also be changed to use isscalar instead of try ... else? Either at __init__ set it to array or None or check it here.

[eendebakpt]

I decided to convert scalar values to a list in the __init__. It takes a bit extra memory, but I think it is rare for the frequency to be so large this causes problems.

Note that the z0_line, etc. are public attributes so this slightly changes the public behavior of the TRL.