Avoid NaN standard deviation from roundoff in Tally.std_dev

[GuySten]

The batch variance is formed as sum_sq/n - mean**2, a difference of accumulated sums. When every realization scores nearly the same value the true variance is zero and rounding can leave the difference just below it, so the square root produces NaN and a RuntimeWarning. Because get_pandas_dataframe() ends with df.dropna(axis=1), a single NaN silently removes the whole "std. dev." column from the dataframe.

Clamp the variance at zero, as mean_stdev() in src/output.cpp already does and as keff_std is guarded in src/eigenvalue.cpp.

Handle a single realization separately, where n - 1 is a division by zero. The standard deviation stays NaN for bins that scored, as it was, but the user now gets a statement of the problem rather than numpy's "invalid value encountered in divide"; bins that never scored stay at zero and stay silent.

Fixes #2408

Checklist

• I have performed a self-review of my own code
• I have run clang-format (version 18) on any C++ source files (if applicable)
• I have followed the style guidelines for Python source files (if applicable)
• I have made corresponding changes to the documentation (if applicable)
• I have added tests that prove my fix is effective or that my feature works (if applicable)