[Code scan] block_avg crashes when no full block remains

Source: Codex global repository scan of deepmodeling/dpti at commit b719828e7eeb571bd26411430197cd74ea78e38c.
Project: https://github.com/orgs/deepmodeling/projects/27

Problem
`block_avg()` assumes at least one full block survives after `skip`. If the input after skipping is empty, `data_chunks[-1]` raises `IndexError`; if it has fewer samples than `block_size`, the only partial block is dropped and NumPy later raises an axis error.

Code references
https://github.com/deepmodeling/dpti/blob/b719828e7eeb571bd26411430197cd74ea78e38c/dpti/lib/utils.py#L105
https://github.com/deepmodeling/dpti/blob/b719828e7eeb571bd26411430197cd74ea78e38c/dpti/lib/utils.py#L112
https://github.com/deepmodeling/dpti/blob/b719828e7eeb571bd26411430197cd74ea78e38c/dpti/lib/utils.py#L120

Reproduction
Run:

```python
import numpy as np
from dpti.lib.utils import block_avg
block_avg(np.arange(5), block_size=10)
```

Expected result
The function should either handle the partial block intentionally or raise a clear validation error explaining that no complete block is available.

Start in dpti/lib/utils.py at block_avg(), especially the referenced lines, and run the provided np.arange(5) reproduction. Decide whether the partial block should be handled or rejected with a clear validation error, then add coverage for empty-after-skip and shorter-than-block-size inputs. Done means these cases no longer produce IndexError or an unclear NumPy axis error.

Written by the indexing model from the issue text.