[antonsoo]

Fixes #119. eofsAsCovariance() unweights a view of the solver's stored data in place, so each call divides the data by its weights again. Repeating the same query can therefore return a different map:

import numpy as np
from eofs.standard import Eof

solver = Eof(
    np.array([[1., 0.], [2., 1.], [3., 3.], [4., 2.]]),
    weights=np.array([.5, 1.]),
)
first = solver.eofsAsCovariance(neofs=1)
second = solver.eofsAsCovariance(neofs=1)
print(np.allclose(first, second))  # Before: False; after: True

Copy the data view when weights are present before dividing it. This preserves the first query's result and leaves the unweighted path unchanged. Add repeated-query checks for all three PC scalings to the existing solver-interface tests.

The configured suite passes with Iris, xarray and Dask on Python 3.12. It also passes on Python 3.8, with the unavailable optional Iris cases skipped. The configured pre-commit checks pass. Repeated queries on both bundled height/SST examples match independent covariance products; first-query results, complete unweighted controls and output metadata remain exact.