TokenClassificationModelWithSeq2SeqEncoderAndCrf: concatenate sequence segments from same batch

[ArneBinder]

TODO: explain approach

usage:

• model config: wrap the seq2seq encoder in concatenate_sequences (use the usual seq2seq config as value of module)
• taskmodule config: set inputs_key_document_indices=seq2seq_encoder_sequence_ids
• don't shuffle data during training!

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 48.38710% with 16 lines in your changes missing coverage. Please review.

Project coverage is 95.17%. Comparing base (201a022) to head (a914614).

Files with missing lines	Patch %	Lines
...c/pie_modules/models/components/seq2seq_encoder.py	30.00%	14 Missing ⚠️
...labeled_span_extraction_by_token_classification.py	77.77%	2 Missing ⚠️

❌ Your patch status has failed because the patch coverage (48.38%) is below the target coverage (90.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #159      +/-   ##
==========================================
- Coverage   95.44%   95.17%   -0.28%     
==========================================
  Files          61       61              
  Lines        5313     5342      +29     
==========================================
+ Hits         5071     5084      +13     
- Misses        242      258      +16     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.