added center_scale_norm

[nkern]

Added center_scale_norm that centers and then normalizes by absolute deviation. Need to keep data centered to train diffusion models with a mean-zero noise scheduler

[JiamingPan]

Two small questions:

1. The function center_scale_norm still says median/absolute deviation, but the code uses mean/max-abs. Should we update the docstring?
2. center_scale_norm currently mutates x in-place with x -= avg and x /= x.abs().max(). Since load_data can use tensors backed by NumPy memmaps, should this be out-of-place, e.g. x = x - x.mean(); x = x / x.abs().max(), to avoid the non-writable-array warning?

[nkern]

great points, will fix this, thanks!

[nkern]

@JiamingPan lmk if you have any other concerns! if not I will merge it