Add depth supervision to the semantic-nerfw model

[yimingzhou1]

This PR integrates depth supervision to the existing semantic-nerfw model.

• Added a include_depth flag to the semantic-nerfw model
• Added (optional) depth information to SemanticDataset
• Added semantic_filenames to the nerfstudio_dataparser
• Updated configurations for semantic-nerfw in method_configs.py

Note: The sitcom3d_dataparser is removed in method_configs.py to pass the check. Looks like it is more unified to have the semantics info included in the nerfstudio_dataparser just like depth info.

Trained the model on Replica Dataset with 2 views of a single scene using both semantics and depth supervision, the results of rendered rgb, depth and semantics of two other adjacent views in the eval set are shown below (left: groundtruth, right: rendered result)

view1 (RGB)

view1 (depth)

view1 (semantics)

view2 (RGB)

view2 (depth)

view2 (semantics)