Add upsampling parameter to Segformer

[thierrypin]

I did this locally and it works perfectly well, but I legally cannot make a pull request. It is simple and easy.

On file segmentation_models_pytorch/decoders/segformer/model.py:

Line 60: Add upsampling: int = 4,, after activation param and before aux_params.
Line 85: swap the parameter value from 4 to upsampling.

This is all. I've done it locally and I think the community would benefit.

Thank you

[qubvel]

Sounds good to me! Will add a tag for contribution form community