Spatial complexity reduction for the run pipeline (partial)

[MoritzSchloesser]

Partial solution to #705 (spatial complexity reduction pipeline
integration — see docs_notes/issue_spatial_complexity_reduction_pipeline_integration.md
for the full issue writeup).

Summary

Adds spatial complexity reduction to the eDisGo run pipeline, as a
counterpart to the existing temporal complexity reduction
(select_timesteps). Spatial reduction merges nearby buses into
representative buses to shrink the grid the OPF runs on; reinforcement
still runs on the full topology.

• New spatial_reduce / spatial_restore tasks bracketing optimize.
  spatial_reduce stashes a deepcopy of the full grid and spatially
  reduces the working object; spatial_restore writes the optimized
  flexible-component dispatch (charging points, heat pumps, DSM loads,
  storage) back onto the stashed full grid and recomputes reactive power.
  Both are no-ops when spatial_reduction.enabled is false.
• New core function apply_reduced_results_to_full_grid +
  EDisGo.map_reduced_results_to_full_grid, mirroring the existing
  spatial_complexity_reduction/EDisGo.spatial_complexity_reduction
  pairing.
• Validator integration via the existing requires/provides capability
  system (no new validator machinery): spatial_reduce provides
  reduced_grid, optimize additionally provides optimized_dispatch,
  spatial_restore requires both.
• New top-level spatial_reduction: config block, mirroring
  timeseries_selection: exactly — same shape, same eGo injection pattern
  (global default + per-grid override).
• New preset uc6_spatial_reduction.yaml and eGo example
  scenario_setting_uc6_example.json.
• eGo: _build_run_edisgo_config injects spatial_reduction/
  spatial_reduction_per_grid the same way it already injects
  timeseries_selection.

This PR covers aggregation_mode=False only. aggregation_mode=True
has a known, documented gap unrelated to this PR's own logic — see
"Known limitation" below and the linked issue.

Testing

• tests/tools/test_spatial_complexity_reduction.py::TestApplyReducedResultsToFullGrid
  — by-name write-back, disaggregation-sum exactness, the numpy-array-input
  regression, the time-index-mismatch guard, and reactive-power recompute,
  all against stub OPF results (no Julia/Gurobi dependency).
• eGo/tests/tools/test_edisgo_integration.py — the
  spatial_reduction/spatial_reduction_per_grid injection logic.
• Verified end-to-end against a real grid (32377), both directly through
  eDisGo (run_example_06.py) and through eGo
  (scenario_setting_uc6_example.json): spatial_reduce → optimize (Gurobi) → spatial_restore → reinforce → save all completed successfully.
• analyse_spatial_reduction.ipynb (not part of this PR, kept local) drives
  the pipeline task-by-task and confirms the reduced grid's OPF output
  exactly matches the full grid's post-restore value for every flexible
  component (0/388 mismatches on the test run).

Known limitation (tracked, not fixed in this PR)

aggregation_mode=True merges loads at the same bus and correctly
aggregates loads_active_power, but does not aggregate
electromobility.flexibility_bands — the OPF's charging-point constraint
builder reads that separately, keyed by charging-point name, so a merged
representative has no upper_power entry and optimize raises KeyError
before spatial_restore ever runs. Only manifests with aggregation_mode: true and 2+ charging points sharing a bus. Full writeup with the exact
traceback and suggested fix in
docs_notes/issue_aggregation_mode_flexibility_bands_not_aggregated.md.
aggregation_mode: false is the default in the new preset and unaffected.

Also included

• Fix: build_flexibility_bands now trims electromobility.flexibility_bands
  to the active time index right after building it — it previously only
  resampled to the active frequency, not the active date range, so a
  manual/short select_timesteps window left bands spanning the full
  underlying data range with nothing ever trimming them down. Unrelated to
  spatial reduction directly, but was blocking end-to-end testing of it
  (see docs_notes/issue_temporal_reduction_flexibility_bands.md).
• Fix: flexible_* or [] in apply_reduced_results_to_full_grid raised
  ValueError on numpy-array input (task_optimize derives
  flexible_loads as edisgo.dsm.p_min.columns.values, not a list) —
  found on the first live end-to-end run.

Not in this PR

• aggregation_mode=True fix (tracked issue, linked above).
• Pipeline/task-level integration tests for spatial_reduce/
  spatial_restore (current scope is core-function-only, by design — see
  docs_notes/spatial_reduction_grilling_session.md).
• Two ADR candidates flagged during design ("two bracketing tasks, restore
  optional"; "disaggregation by pre-OPF flexibility envelope") — not yet
  written up.