Benchmarks to optimize prediction speed

fastai verison: 1.0.55.dev0

Classification

CPU:

Method	~ sec/image
`.predict` iteration	0.24
`.pred_batch`	0.17
`.pred_batch` + `apply_tfms`	0.18

GPU (p2.xlarge):

Method	~ sec/image
`.predict` iteration	0.019
`.pred_batch`	0.0061
`.pred_batch` + `apply_tfms`	0.0073

GPU (p3.2xlarge):

Method	~ sec/image
`.predict` iteration	0.013
`.pred_batch`	0.0034
`.pred_batch` + `apply_tfms`	0.0038

Segmentation

CPU:

Method	~ sec/image
`.predict` iteration	8.3
`.pred_batch`	7.7
`.pred_batch` + `apply_tfms`	8.2

GPU (p2.xlarge):

Method	~ sec/image
`.predict` iteration	0.45
`.pred_batch`	0.29
`.pred_batch` + `apply_tfms`	0.29

GPU (p3.2xlarge):

Method	~ sec/image
`.predict` iteration	0.048
`.pred_batch`	0.032
`.pred_batch` + `apply_tfms`	0.033