Minor bug fixes that seemed necessary to get v2.0 to run

[jhennawi]

In the process of trying to get v2.0 to run, I encountered a few bugs, which this pull request addresses.

[bd-j]

Thanks, I think this has overlap with some of the fixes in #379 so I'll try to fix any conflicts; I may end up not merging this but rather importing some of the changes into #379.

[bd-j]

@jhennawi I have merged #379 which obviated some of these fixes (the var-covar typo was fixed, and the predict_intrinsic() method was removed in favor of the quantities() method). I added a commit (fe3322d) that moved the response caching to before MLE line fitting, and made sure the Spectrum.compute_response method returned a vector of same length as the input spectrum. Thanks much for flagging these problems! Please let me know if you encounter continued issues with v2.0 and MLE line fitting.