Automatically download embeddings

[kermitt2]

As part of the embeddings and model management in DeLFT, download on demand word embeddings, contextualized embeddings and transformer models.

Thanks to @de-code, the functionality to automatically download embeddings is already available !

[de-code]

Just to add that to that, I've also added preloading of embedding as part of the Docker entrypoint via an environment variable (PRELOAD_EMBEDDING).

I did try it with xz (LZMA) compressed files but found it took too long to decompress. I then reverted to using the decompressed mdb (LMDB cache) even though it was more than twice the size. But that probably depends on the setup and network connection.