Replace deprecated weights kwarg for torchvision pretrained models

[shaikh58]

PyTorch has deprecated the weights kwarg for its pretrained models API. It has been replaced with 'weights=....'. Currently our model pipelines use timm as the backend which works with the 'pretrained' flag, but this will break if 'torch' is selected as the backend.

From torchvision docs (https://pytorch.org/vision/2.0/models.html):

resnet50(weights=ResNet50_Weights.IMAGENET1K_V1)
resnet50(weights="IMAGENET1K_V1")
resnet50(pretrained=True)  # deprecated
resnet50(True)  # deprecated

Requires minor change to config file structure in model.encoder_cfg to add the weights parameter, and minor changes to visual_encoder.py to not pass the pretrained param to the constructor, but rather the weights param. The 'weights param should not be passed to timm constructors. The config names should be changed to make it more clear that weights is for PyTorch and pretrained is for timm. E.g. timm_pretrained and torch_weights