[Code scan] einstein CLI passes the wrong type to free_energy

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

Problem
The `dpti.einstein` CLI loads the input JSON into a dict and passes that dict to `free_energy()`. The `free_energy()` API expects a job directory path and immediately calls `os.path.join(job, 'in.json')`, so the CLI path is broken.

Code references
https://github.com/deepmodeling/dpti/blob/b719828e7eeb571bd26411430197cd74ea78e38c/dpti/einstein.py#L58
https://github.com/deepmodeling/dpti/blob/b719828e7eeb571bd26411430197cd74ea78e38c/dpti/einstein.py#L192
https://github.com/deepmodeling/dpti/blob/b719828e7eeb571bd26411430197cd74ea78e38c/dpti/einstein.py#L193

Reproduction
Run:

```bash
python -m dpti.einstein some_input.json
```

Expected result
The CLI should either accept a job directory and pass it through to `free_energy()`, or the API should be changed to accept the loaded JSON consistently.

Start in dpti/einstein.py at the CLI handling around lines 58, 192, and 193, then inspect free_energy() to confirm its expected argument. Reproduce the failure with `python -m dpti.einstein some_input.json`. Done means the CLI and API agree on the input type and this command no longer passes an incompatible value.

Written by the indexing model from the issue text.