remove pie-modules

[ArneBinder]

... in favor of pie-documents (documents, annotations, etc) and pytorch-ie (models, taskmodules, and tokenize_document).

Details and additional changes:

• update dependencies:
  • pytorch-ie: >=0.33.0
  • pie-datasets: >=0.11.0
  • pie-modules: remove
• Update dataset revisions in configs to latest versions on the HF hub (these include this: use pie-documents (in dataset scripts) pie-datasets#209). This is important since older versions do not work along with pie-documents.
• Update the local version of the conll2003 dataset script with the latest version from pie-datasets.
• Use pytorch_ie.PyTorchIEPipeline instead of pytorch_ie.AutoPipeline in configs.
• Use generic pie_core.AnnotationPipeline instead of pytorch_ie.Pipeline in main scripts (just type annotations).
• Annotate the example model implementation (MyTransformerTokenClassificationModel) with @Model.register() instead of @PyTorchIEModel.register(). This is important to get model loading working from now on.
• Use serialize_document_type from pie_core.utils.hydra (instead of from pytorch_ie.utils.hydra).
• Replace deprecated AnnotationList with AnnotationLayer.
• Since pytorch-ie relies now on Python>=3.10 (see switch to 3.10 as min Python version pytorch-ie#500), this also changes the min Python version accordingly.
• Update pie-core to 0.3.0 (just in the lock file).

requires:

• pie-datasets >=0.11.0
• dont use .convert_dataset in predict script #214

TODO:

• use pytorch-ie 0.33.0 when available. requires:
  • add taskmodules and models from pie-modules pytorch-ie#498
  • register all models with @pie_core.Model pytorch-ie#503
  • remove backwards compatibility pytorch-ie#507
  • fix failing test test_pipeline_with_documents pytorch-ie#506
  • create pytorch-ie release v0.33.0

Note: I triple-checked that everything is running (slow tests), including drugprot experiments.