Add `--cache val` (#6663)

New `--cache val` argument will cache validation set only into RAM. Should help multi-GPU training speeds without consuming as much RAM as full `--cache ram`.

train.py

 
     # Trainloader
     train_loader, dataset = create_dataloader(train_path, imgsz, batch_size // WORLD_SIZE, gs, single_cls,
-                                              hyp=hyp, augment=True, cache=opt.cache, rect=opt.rect, rank=LOCAL_RANK,
-                                              workers=workers, image_weights=opt.image_weights, quad=opt.quad,
+                                              hyp=hyp, augment=True, cache=None if opt.cache == 'val' else opt.cache,
+                                              rect=opt.rect, rank=LOCAL_RANK, workers=workers,
+                                              image_weights=opt.image_weights, quad=opt.quad,
                                               prefix=colorstr('train: '), shuffle=True)
     mlc = int(np.concatenate(dataset.labels, 0)[:, 0].max())  # max label class
     nb = len(train_loader)  # number of batches
     # Process 0
     if RANK in [-1, 0]:
         val_loader = create_dataloader(val_path, imgsz, batch_size // WORLD_SIZE * 2, gs, single_cls,
-                                       hyp=hyp, cache=None if noval else opt.cache, rect=True, rank=-1,
-                                       workers=workers * 2, pad=0.5,
+                                       hyp=hyp, cache=None if noval else opt.cache,
+                                       rect=True, rank=-1, workers=workers * 2, pad=0.5,
                                        prefix=colorstr('val: '))[0]
 
         if not resume: