Follows discussion in #1454

Changes proposed in this Pull Request

When using optimize_with_rolling_horizon with linearized_unit_commitment and ramp limits defined for committable generators, the optimization fails with "KeyError: "not all values found in index 'snapshot'". Here is a minimalistic example to reproduce h/t @thomgeo

import pypsa
n = pypsa.examples.scigrid_de()

disp = ["Gas", "Hard Coal", "Waste", "Brown Coal", "Oil", "Nuclear"]
n.generators.loc[n.generators.carrier.isin(disp), "committable"] = True
n.generators.loc[n.generators.carrier.isin(disp), "ramp_limit_up"] = 0.5

n.optimize.optimize_with_rolling_horizon(
    solver_name="gurobi", linearized_unit_commitment=True, horizon=12
)

(FYI @thomgeo though it is not related to ramp values >1 per se)

The error occurs in the ramp limit constraints for committable components when trying to adjust the RHS for the first snapshot of each rolling horizon window. The problematic code is here:

If we step in with debugger in this line for an example above: the .item(0) method returns the snapshot coordinate's int64 representation instead of a datetime object.

{"snapshot": rhs.coords["snapshot"].item(0)}
{'snapshot': 1293883200000000000}

Later this integer is used to index into a DatetimeIndex, and pandas complains.

For the fix I use xarray position-based .isel() indexing instead for two places in the code (ramp up/down). I'll add some tests too later tonight.

Checklist

• Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in docs.
• Unit tests for new features were added (if applicable).
• A note for the release notes docs/release-notes.md of the upcoming release is included.
• I consent to the release of this PR's code under the MIT license.