Add EfficientLoFTR model

[sbucaille]

Model description

EfficientLoFTR is an image matching model performing dense matching (contrary to SuperPoint + SuperGlue).
It is a variant to LoFTR that runs in realtime (at most 40ms for inference).
The base model performances are ok, but with the soon release of MatchAnything version of EfficientLoFTR, we should have an even better model.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

https://github.com/zju3dv/EfficientLoFTR