Guard short token datasets in ANE and dynamic training paths

[TastyHeadphones]

Summary

• add a token dataset length guard in training/train_large_ane.m
• add the same guard in training/training_dynamic/train.m
• fail early with a clear error when the dataset is too short for one (input,target) training window

Why

Both paths sample with:

max_pos = n_tokens - SEQ - 1

When n_tokens <= SEQ + 1, this unsigned subtraction can underflow, producing a huge random range and potentially leading to out-of-bounds reads.

Validation

• make -C training train_large_ane
• make -C training/training_dynamic train