Read stanford configs

[NohTow]

Right now, we are using default values for doc/query length, markers and attending to expansion tokens when reading stanford models.
This causes some issues as highlighted in #85 and also requires the user to specify a lot of information when using a model that is not using default values, as can be seen with the loading of Jina-ColBERT.

This PR simply add the reading process of the artifact.metadata file of Stanford NLP models and read markers, lengths and attend to expansion tokens values.

As usual, we override those if the user feed values to the init of the model. Also changed the attend_to_expansion_tokens parameter to match the other (None by default and override at the end).

[NohTow]

I think it should be ok @raphaelsty
Could you please review to make sure the behavior is in line with what's expected? I did some tests, but I might have missed something
Note that it is based on #87, so we should merge this one first

[sam-hey]

It seems that my reviews are in Pending state, so they can't be seen at the moment. I have just one small regrade:

I wouldn’t categorize this as a warning, as it’s expected behavior. It’s merely informing you that the StanfordNLP model has successfully loaded the weights.

pylate/pylate/models/colbert.py

Line 268 in f0899ab

logger.warning("Loaded the weights from Stanford NLP model.")

Similarly, I believe there should be a clear distinction between informational events and actual warnings. For instance, if there were an issue loading the file and it fell back to a default, that would merit a warning.

pylate/pylate/models/colbert.py

Lines 294 to 303 in f0899ab

	logger.warning("Loaded the configuration from Stanford NLP model.")
	except EnvironmentError:
	if self.query_prefix is None:
	self.query_prefix = "[unused0]"
	if self.document_prefix is None:
	self.document_prefix = "[unused1]"
	# We do not set the query/doc length as they'll be set to the default values afterwards. We do it for prefixes as the default from stanford is different from ours
	logger.warning(
	"Could not load the configuration file from Stanford NLP model, using default values."
	)

[NohTow]

That is very fair, I don't know why I started using logger.warning, I think I used it for a proper warning once and then copied it over the whole loading logic, my bad.
I'll clean that up before merging, thanks for the remark!

[NohTow]

@sam-hey I changed some warnings to info
@raphaelsty you can merge after you have double checked the loading/overriding logic

[raphaelsty]

LGTM