More default experiments and sweeps

[alex404]

Going to try and add a set of default */lindsey.yaml configs for sweeping through the Lindsey et al results. Beyond just having those results on hand, it'll serve the purpose of having a set of simulations that's meant to stable. In contrast, the more generic classification.yaml experiment will adapt to enhancements and features to the library.

• [ ] Develop and add */lindsey.yaml files.
• [ ] Bonus points: See if we can implement a "SOTA" (e.g. alexnet) network on the scaled+shifted data, to provide a better performance upper bound on the task.

[fabioseel]

bold move to call alexnet state of the art :D Maybe also a ResNet (or ConvNext as the modernized version), perhaps even pretrained would give a more realistic upper bound

[alex404]

bold move to call alexnet state of the art

That's why I wrote "SOTA" not SOTA ;)