Use real component dtype when writing statistics for complex data

[gshiroma]

This PR updates BaseL2WriterSingleInput to ensure statistics computed from complex-valued data (e.g., GCOV off-diagonal terms) are written using the corresponding real-valued dtype.

[nemo794]

Adding more context to this PR description:

Currently, GCOV off-diagonal complex-valued datasets (e.g., HHHV with complex64 dtype), the statistics attributes (min_real_value, mean_real_value, min_imag_value, etc.) are incorrectly being stored as complex-valued (e.g., complex64) with the actual metric saved to the real component. (See screenshot)

However, for a complex64 dataset (composed of float32 + float32), statistics attributes should be stored as float32, not complex64, since the statistics themselves are real-valued scalars. This was inconsistent with RSLC and GSLC products, where statistics for complex-valued datasets are correctly stored as float-valued.

This PR addresses that inconsistency.

[nemo794]

Thanks @gshiroma !

I'm pretty sure you tested this PR awhile back. Please attach a screenshot of the output product, as a sanity check to confirm that these fields are being populated correctly now?

See #306 (comment) for an example screenshot of what the metadata looked like previously. Thanks!