Update token_classification notebook to use evaluate.load() instead of load_metric() from Datasets library (removed)

[BibWiss]

trafficstars

What does this PR do?

This PR updates the token_classification notebook in terms of evaluation metric loading. As of now, it uses the load_metric function from the Datasets library - however, this one has been removed in [email protected] (see also: https://discuss.huggingface.co/t/cant-import-load-metric-from-datasets/107524/2). Therefore, it has been changed to evaluate.load() (and explanatory descriptions have been updated accordingly).

Who can review?

Feel free to tag members/contributors who may be interested in your PR. @sgugger