Add TimeSeries.argmax() and argmin() aggregation methods

[jaideeppyne]

Summary

Adds TimeSeries.argmax() and TimeSeries.argmin(), which return the integer position of the maximum/minimum along a given axis, following numpy.argmax/numpy.argmin semantics.

This addresses the second half of #2696, which notes:

There also does not seem to be a TimeSeries equivalent of Series.idxmax or np.ndarray.argmax.

Rather than changing the existing max()/min() behaviour (which the issue observes is kept for consistency with mean(axis=0), median(axis=0), … and would be a breaking change), this adds the missing positional reducers alongside them.

With axis=0, the returned position maps straight back to a timestamp:

pos = series.argmax(axis=0)          # position of each component's max over time
ts_of_max = series.time_index[int(pos.values()[0, 0])]

Details

• argmax(axis=2) / argmin(axis=2) mirror the existing max()/min() implementation exactly, reusing _get_agg_dims for the output time index and component naming (components_argmax / components_argmin when reducing over components).
• Works over all three axes (time, components, samples), consistent with the other reducers.
• Non-breaking: purely additive.

Testing

• Added test_argmin / test_argmax to TestSimpleStatistics, checking both that the positions equal numpy.argmin/argmax and that each returned position actually points back to the true extremum (via np.take_along_axis) across all three axes.
• New tests pass; the full TestSimpleStatistics suite is green; ruff check and ruff format --check pass.
• Added a CHANGELOG.md entry under Improved.

[dennisbader]

Thanks for the PR @jaideeppyne. This is already being worked on in #3115, so I'm closing this one in favor of the other. Feel free to add your comments / review there.