Fix spectra does not deal correctly with set_reference

[RadixSeven]

When doing a set reference, fix_spectra copies the values to the right x indices in the new spectrum but has two problems

1. y values that will be moved to x's outside the range after the shift are discarded. This can conceivably lose precious data that doesn't need to be lost.
2. y values from x's that originally lie outside the range and that the shift would bring into the range are left unchanged - giving realistic looking garbage

The second problem is the worst. Imagine that a spectrum needs to be shifted 7 array indices. Then the first seven and second seven indices will hold duplicate y values. This can throw off error standard-deviation estimation which uses the first 30 points.

A proposed fix

Initially, keep a different set of x-values for each spectrum. We can create this new set by adding the shift amount to each x value.

Then take the intersection of the ranges covered by the resulting x values as the new range. Choose the union for all spectra of all the x-values that fall in this range. These are the new x-values. For each spectrum, linearly interpolate the y-values at x values that were not directly measured and copy the rest.

[RadixSeven]

To reproduce

1. Run fix_spectra
2. Open smaller_anit.zip
3. Set reference making 6.08-6.05 (Allantonin) go to 0.

Result

The resulting spectrum will obviously be the result of pasting a copy of the upper half of the spectrum on top of the lower half.

What was expected

The resulting spectrum should be almost unchanged with the exception of a few points lost on the edges.