Make some "science use case" tests for the analysis tools

Right now some of the unit tests do things that aren't always that meaningful scientifically, like measuring the centroid of multiple lines or the like. That is often fine/good for unit tests, but we should make an effort to include some "is the science right" tests. In particular I'm thinking of:

* Build a fake spectrum *known* line properties and add noise
* Try the continuum subtraction algorithms
* Measure properties like EQW, flux, FWHM, etc
* Compare to the *expected* results for those given the *noiseless* line properties. These can't be exact checks for equality because of the noise, but something that's a "good enough" (e.g., is the centroid right to within a pixel or two).

(Note: some of the tests I'm referencing are in in-process PRs, so right now this is more of a "do it later" issue. Should become more relevant once those PRs are merged, though.)

Start by reviewing the existing unit tests for the analysis tools and the in-process PRs referenced in the issue. Build a noisy synthetic spectrum with known line properties, exercise continuum subtraction and measurements such as EQW, flux, and FWHM, and verify results against the noiseless expectations using appropriate tolerances.

Written by the indexing model from the issue text.