[Code scan] Keep SIESTA density-matrix keys aligned with filtered blocks

This issue comes from a Codex global repository scan.

## Problem
SIESTA density-matrix parsing filters keys with `block_mask`, but zips those filtered keys with the unfiltered `block` array:

https://github.com/deepmodeling/dftio/blob/c9d128f24a74ef2911e1a28f5640357488beb196/dftio/io/siesta/siesta_parser.py#L530-L535

If an earlier R block is below threshold, later retained keys receive the wrong matrix. This silently corrupts density-matrix blocks.

## Reproduction
With `block_mask = [False, True]`, the first retained key is zipped with `block[0]`, not `block[1]`.

## Suggested fix
Filter values with the same mask before zipping:

```python
DM_blocks_dict[block_index] = dict(zip(keys, block[block_mask]))
```