seq2ribo 0.2.0: corrected models

After publication we found a bug in the data preprocessing: for transcripts on the minus strand, the A-site
profiles were stored in genomic order, i.e. reversed relative to the transcript. This release fixes it and
ships models retrained on the corrected data. UPDATE.md
describes the fix and the results.

• The corrected models are the default (weights/<cell>/). The published models remain available with
  Seq2Ribo(..., version="paper"), which downloads seq2ribo-paper-weights.tar.gz from this release on first use.
• reproduction/ holds the code of the published and of the corrected pipeline.
• The repository history was rewritten to remove superseded checkpoints; existing clones should be re-cloned.

Assets

File	Contents	sha256
seq2ribo-paper-weights.tar.gz (75 MB)	published checkpoints; tar -xzf seq2ribo-paper-weights.tar.gz -C weights creates weights/paper/	924a193517bfc95a75f6dbad83e2af219709a141268833502af6c20ea0ed578d
seq2ribo-corrected-results.tar.gz (648 MB)	every checkpoint and result behind the UPDATE.md tables (ensemble members, sTASEP/TASEP fits, RiboNN and Translatomer checkpoints); extract into reproduction/corrected/	c61771a1027a4a52a98cbcf860f9c90d1634cf4430a9fd205014cc31cbdb6606