[Code scan] Pressure task reports success_rate with the wrong denominator

This issue was found by a Codex global repository scan of tracked non-test files at commit `8c93925cb10b401b2b83c738bd9263fd74474468`.

### Relevant code
https://github.com/deepmodeling/LAMBench/blob/8c93925cb10b401b2b83c738bd9263fd74474468/lambench/tasks/calculator/pressure/pressure.py#L57-L84

### Impact
`num_samples` is incremented only after a successful calculation, while `num_fails` is incremented for failed attempts. The final formula is `(num_samples - num_fails) / num_samples`, which can underreport success and can crash when every sample fails.

For example, with one successful sample and one failed sample, the current formula reports `(1 - 1) / 1 = 0.0` instead of the expected `1 / 2 = 0.5`.

### Suggested fix
Track total attempts separately from successful samples. Compute `success_rate = num_successes / total_attempts`, and return controlled `NaN` or `None` metrics plus `0.0` success when there are no successful samples.

Start in lambench/tasks/calculator/pressure/pressure.py at lines 57-84 and trace how successful samples, failed attempts, and the final metric are counted. Exercise the pressure task with mixed and all-failing samples, then update the calculation so success uses total attempts and no-success cases return controlled metrics with a 0.0 success rate.

Written by the indexing model from the issue text.