FRML-153 Migrate const rotated dataset spec as variant in presets

[Eve-ning]

Previously, when we create a static evaluation dataset (i.e. 8 unique orients) we need to invoke the dataset as:

ds=FRDCDatasetStaticEval(
    "chestnut_nature_park",
    "20210510",
    "90deg43m85pct255deg",
...

This was clunky and flaky for several reasons:

1. We had to manually specify the exact string path to the dataset
2. Static Eval was entirely decoupled from the preset ecosys
3. It was inconsistent with the preset syntax we used for train and validation

Now, we adapt the code to use the following, which is cleaner

ds=ds.chestnut_20210510_43m.const_rotated(
    transform=const_weak_aug(im_size),
    transform_scale=train_lab_ds.x_scaler,
),

This follows the same motif as .labelled, .unlabelled when we invoke any dataset preset.

Note

One can technically change any FRDCDataset to FRDCUnlabelledDataset or FRDCStaticRotatedDataset by simply changing the __class__ attribute to force it to use another interface

Major Changes

• Moved FRDCStaticEvalDataset to presets.py as a dataset variant
• Updated training scripts to use new dataset variant