Batch img refactor

[mrariden]

Add back in support for batches of images as numpy arrays (eg [N, H, W, C]).

Timing results

Using a random numpy array of (50, 1024, 1024, 3) on a RTX 6000, averaged over 10 calls to .eval()

batch_size	time (s)
1	10.7
5	9.06
10	8.79
20	9.08
40	8.94
60	8.85

Using a random numpy array of (50, 50, 50, 3) on a RTX 6000, averaged over 10 calls to .eval()

batch_size	time (ms)
1	903
10	729
60	713
100	726

Using a stack of 512, 240x240 images on a RTX 6000, no averaging:

batch_size	forward pass time (s)	total net time (s)	total .eval() time (s)
1	0.015	14.0	30.7
4	0.033	5.43	21.0
8	0.0615	5.02	21.1
16	0.120	4.84	20.6
32	0.244	4.82	20.7

[mrariden]

resolves #1372