Closes # (if applicable): n/a

Changes proposed in this Pull Request

I was doing some basic profiling, and it looked like a decent portion of the setup time was being spent in this part of the code:

In the case the input values is None, since we set values_ = n_attr, the subset check is redundant and we can skip that block.

(more timing information below)

Checklist

• Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in docs. (n/a it's a one-line change with a comment)
• Unit tests for new features were added (if applicable). (n/a)
• 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.

I used this to measure the total time:

from pypsa.networks import Network
import time

PATH = 'examples/networks/model-energy/model-energy.nc'

def main():

    start_time = time.monotonic()
    n = Network(PATH)
    n.optimize(solver_name="highs", log_to_console=False)
    
    end_time = time.monotonic()
    print(f"elapsed: {end_time - start_time}")

if __name__ == "__main__":
    main()

Running before and after the change 5 times each, the best times were:
before: elapsed: 7.80225883400999
after: elapsed: 7.512157290999312

So it's not a huge difference overall, but since most of that time is spent in the solver, I'd say it's still worthwhile.

To profile where the time was going, I used the technique described here: https://www.youtube.com/watch?v=ey_P64E34g0
In order to exclude the solver time from the profile, I used this to save a pstats file:

pr = cProfile.Profile()
pr.enable()
def stop_profile(n, snapshots):
    pr.disable()
n.optimize(solver_name="highs", log_to_console=False, extra_functionality=stop_profile)
pr.dump_stats("model-energy.optimize.pstats")

and converted to an svg with

gprof2dot model-energy.optimize.pstats -n 5 | dot -Tsvg -o model-energy.optimize.svg

That produced a graph that looked like this:

So roughly half of the non-solver time was spent in as_index(), in particular the all block (although take the actual times with a grain of salt, because the profiler introduces some overhead)