Add a function terratorch.create_model()

[blumenstiel]

Is your feature request related to a problem? Please describe.
Currently, TerraTorch has no user-friedly way to create full models. Also, directly calling the registry does not look very nice IMHO:

model = BACKBONE_REGISTRY.build("prithvi_vit_100")

Describe the solution you'd like
It would be good if we add a funktion terratorch.create_model() similar to timm:

model = timm.create_model('resnet34')

The default could be a simple wrapper around BACKBONE_REGISTRY.build. If a user passes additional parameters like task='classification', TerraTorch could automatically build a suitable model by selecting good default necks, decoders, and heads. Similar, passing task='segmentation' could return a PixelWiseModel. E.g.:

model = terratorch.create_model(
    'prithvi_eo_v2_300', 
    pretrained=True, 
    task='segmentation', 
    num_classes=2,
    decoder='UperNetDecoder'
)

Challenges
This might require that registered backbones come with a attribute for suitable default necks. E.g. we could add default necks to our registered models to always return 4 (intermediate) layers with reshaped 2D patch embeddings. This would help a lot of users independed of this feature request. Of course, users can overwrite these default necks.