Benchmark mantid-like second-order with textbook implementation

The Mantid-like second-order calculation deviates from a "textbook" implementation in two significant ways:

- by iterating over individual q-points rather than q-point pairs (with significant performance implications)
- applying an extra factor of 0.5

It would be good to compare these over a range of materials and against experimental data, to understand the success of this method and whether it would be worth switching for general (or only for "best-quality") calculations.

It would make sense to do this after #33 so that isotropic self-convolution is also part of the comparison; this might be the "draft-quality" option.

No files, tests, materials, or experimental datasets are named. Start by locating the Mantid-like second-order calculation and review the dependency on #33, then define comparisons of q-point versus q-point-pair iteration and the extra 0.5 factor across materials; done means the results are compared with experimental data and support a decision on which method to use.

Written by the indexing model from the issue text.