Describe the bug

AptaTrans.__init__() raises AssertionError when in_dim is not divisible by n_heads. This is user-facing input validation, not an internal invariant check — the correct exception type is ValueError.

Using AssertionError for input validation is an anti-pattern because:

• AssertionError signals broken internal assumptions, not bad user input
• Users may catch ValueError but not AssertionError in their error handling
• It's inconsistent with Python and scikit-learn conventions

To Reproduce

from pyaptamer.aptatrans import AptaTrans, EncoderPredictorConfig

apta = EncoderPredictorConfig(128, 16, max_len=128)
prot = EncoderPredictorConfig(128, 16, max_len=128)

# Raises AssertionError instead of ValueError
model = AptaTrans(apta, prot, in_dim=128, n_heads=3)
# AssertionError: Input dimension 128 must be divisible by number of heads 3.

Expected behavior

Should raise ValueError with the same message, since this is validating a user-provided parameter value.

Additional context

• The docstring also contains a typo: AssertionError (missing the 's' — though AssertionError is actually valid Python, the standard spelling is AssertionError)
• Scikit-learn and PyTorch conventions both use ValueError for parameter validation

Versions

pyaptamer 0.1.0a1