Improved huggingface batch logic

[ajsalow]

Addresses issues described in #1249 where batch size greater than 0 causes some hugging face models to fail loading.

[adriangonz]

Changes look good! Thanks for taking the time to contribute this patch @ajsalow.

Before merging, could you add a test covering the new behaviour? Besides that, there also seem to be some conflicts with latest master.

[ajsalow]

Thanks for taking a look @adriangonz. I've added some test coverage, let me know if anything else is needed.

[adriangonz]

Looks good! Thanks for adding that test @ajsalow 👍

Once tests are green, we should be able to merge this one.

[adriangonz]

Hey @ajsalow ,

Just had a look at the test failures. I think from those, the lint failure is probably the key one - everything else seems to be just flaky. So once you get a chance to address the linter error, we should be good to merge this one:

./runtimes/huggingface/mlserver_huggingface/common.py:72:89: E501 line too long (120 > 88 characters)

[ajsalow]

@adriangonz apologies, I incorrectly assumed linting was using pylint. make lint is now passing for me locally.