[ENH] AptaTrans test coverage

[NennoMP]

Following #177, we should review test coverage of AptaTrans to make sure everything works correctly.

Below, the test coverage for AptaTrans related modules.

=============================== tests coverage ================================
_______________ coverage: platform win32, python 3.12.9-final-0 _______________

Name                                                            Stmts   Miss Branch BrPart  Cover
-------------------------------------------------------------------------------------------------
pyaptamer\aptatrans\__init__.py                                     6      0      0      0   100%
pyaptamer\aptatrans\_model.py                                      80     10     14      2    85%
pyaptamer\aptatrans\_model_lightning.py                            34      0      0      0   100%
pyaptamer\aptatrans\_pipeline.py                                   42      0      6      1    98%
pyaptamer\aptatrans\layers\__init__.py                              3      0      0      0   100%
pyaptamer\aptatrans\layers\_convolutional.py                       24      0      4      0   100%
pyaptamer\aptatrans\layers\_encoder.py                             38      0      2      1    98%
pyaptamer\aptatrans\layers\_interaction_map.py                     14      0      0      0   100%
pyaptamer\aptatrans\layers\tests\__init__.py                        1      0      0      0   100%
pyaptamer\aptatrans\layers\tests\test_all_aptatrans_layers.py      78      0      2      0   100%
pyaptamer\aptatrans\tests\__init__.py                               1      0      0      0   100%
pyaptamer\aptatrans\tests\test_aptatrans.py                       140      4      0      0    97%
pyaptamer\aptatrans\tests\test_aptatrans_lightning.py              66      0      0      0   100%

Coverage looks good overall. Only _model.py is relatively low, however the branch not being tested is the one related to loading pretraining weights, not crucial.

As I mentioned in #178, the bug on shape mismatch was caused by introducing the attention mask in the encoders. This wasn't something we could have catched via test as it's, rather than a bug, was an non-conventional way of doing things (i.e., not using an attention mask).

FYI @fkiraly

[fkiraly]

can we add a small set of weights, or export then import them?

[NennoMP]

can we add a small set of weights, or export then import them?

You mean in order to test the pretrained weights loading branch? I can generate some randoms ones, store them in a file, and use it to test weights' loading.