[Code scan] Handle no-padding pretraining batches in coordinate prediction

This issue is a result of a Codex global repository scan.

## Summary
The pretraining model sets padding_mask to None when a batch has no padding. The coordinate prediction branch later unconditionally calls padding_mask.float(), so batch size 1 or equal-length batches can crash when masked_coord_loss is enabled.

## Code references
https://github.com/deepmodeling/unimol_tools/blob/4596596aa8f73eb462d5cc5a921d79966d0465da/unimol_tools/pretrain/unimol.py#L71-L73
https://github.com/deepmodeling/unimol_tools/blob/4596596aa8f73eb462d5cc5a921d79966d0465da/unimol_tools/pretrain/unimol.py#L100-L116

## Impact
Valid pretraining batches with no padded tokens can raise AttributeError before loss computation. This is especially likely for small datasets, homogeneous molecule sizes, or batch_size=1.

## Suggested fix
Keep a separate token-valid mask before nulling padding_mask for the encoder, or branch in the coordinate head and use an all-ones pair mask when padding_mask is None.

Start in unimol_tools/pretrain/unimol.py at lines 71-73 and 100-116, following how padding_mask is handled before the coordinate prediction loss. Verify the no-padding path with masked_coord_loss enabled, including batch_size=1 or equal-length batches. Done means valid batches without padding reach loss computation without an AttributeError while padded batches retain their existing behavior.

Written by the indexing model from the issue text.