Fix AveragedCrossspectrum for input_counts=False light curves

[VrtxOmega]

Relevant Issue(s)/PR(s)

Fixes #977.

Provide an overview of the implemented solution or the fix and elaborate on the modifications.

AveragedCrossspectrum was passing the private light-curve storage fields (_counts and _counts_err) into the shared time-series cross-spectrum helper. For light curves created with input_counts=False, those private counts arrays can be unset until the public counts/counts_err properties are accessed, so the helper can treat the data like event input and return zero powers.

This updates the light-curve path to pass the public counts and counts_err attributes, matching the existing AveragedPowerspectrum behavior, and adds a regression test for an identical countrate light curve where the averaged cross spectrum should match the averaged power spectrum.

Is there a new dependency introduced by your contribution? If so, please specify.

No.

Any other comments?

Validation run locally:

• .venv/bin/python -m pytest stingray/tests/test_crossspectrum.py::TestAveragedCrossspectrum::test_input_countrate_matches_pds_for_identical_lightcurves -q
• Issue repro before/after: input_counts=False changed from all-zero cross-spectrum powers to matching AveragedPowerspectrum, while input_counts=True remained matching.