[Code scan] Parse ABACUS DeepKS flags before deciding required files

This issue is a result of a Codex global code scan of `deepmodeling/fpop` at commit `b05b337590c31a5237b2dcbd9c0833b841c08cd4`.

Relevant code:
https://github.com/deepmodeling/fpop/blob/b05b337590c31a5237b2dcbd9c0833b841c08cd4/fpop/abacus.py#L399-L424
https://github.com/deepmodeling/fpop/blob/b05b337590c31a5237b2dcbd9c0833b841c08cd4/fpop/abacus.py#L466-L493

Problem:
`AbacusInputs.read_inputf()` stores all INPUT values as strings. `AbacusInputs.write_deepks()` then checks those values directly in boolean conditions:

```python
if self._input.get("deepks_out_labels", False):
need_descriptor = True
if self._input.get("deepks_scf", False):
need_descriptor = True
need_model = True
```

Strings such as `"0"`, `"false"`, and `"False"` are truthy in Python, so an INPUT file that explicitly disables DeepKS can still force descriptor/model files and fail with an assertion.

Minimal reproduction:
```text
INPUT_PARAMETERS
deepks_scf 0
deepks_out_labels 0
```

Constructing `AbacusInputs(INPUT, {"H": "H.upf"})` and calling `write_deepks()` raises `AssertionError` because `"0"` is treated as enabled.

Expected behavior:
DeepKS switches should be parsed with ABACUS-compatible boolean semantics before `write_deepks()` decides whether descriptor or model files are required.