docs: add memory best-practices notebook for large-scale models

[MaykThewessen]

Summary

Adds a notebook documenting memory-efficient patterns for large-scale linopy models, as invited by @FBumann in #599.

Closes #599 (partial — notebook contribution specifically).

Context

The patterns are drawn from production experience with a 200-bus, 1,800-generator PyPSA power system model solved over 8,760 hourly snapshots in 168-hour weekly chunks (~1.4M LP variables and constraints per chunk, ~9 GB steady-state, ~15 GB peak).

Notebook contents (examples/memory-best-practices.ipynb)

1. Understanding linopy's memory layout — formula table for variable/constraint/sparse-matrix storage; the two memory spike points (constraints.flat during matrices.A and COO→CSC conversion; solution unpacking); how to avoid repeated matrix builds
2. Chunked time-horizon solving — the basic pattern; why del network.model + malloc_trim(0) / malloc_zone_pressure_relief() is essential (RSS does not drop without the OS-level trim)
3. LP blueprint reuse across chunks — for models where LP structure is identical across chunks, extract the HiGHS object after chunk 1 and call changeRowsBounds() / changeColsCost() for chunks 2+, skipping linopy rebuild entirely; saves ~50 s and ~9 GB per chunk
4. Efficient RHS and objective updates — numpy broadcast fill vs pandas fillna to avoid (T×N) intermediate copies; sparse RHS push via keys >= 0 masking
5. Memory profiling — tracemalloc and psutil.rss_gb() patterns; estimate_model_memory() helper for sizing RAM before solving
6. Explicit cleanup between solves — matrices.clean_cached_properties() (frees flat_cons cache, typically 2–3 GB); detach solver_model before del; gc.collect() + OS trim
7. Sparse-friendly constraint formulation — filter_missings=True; groupby aggregation vs dense broadcasting; merge=False for per-snapshot constraints

All code examples are from production code paths and include both the recommended pattern and the anti-pattern it replaces.

Files changed

• examples/memory-best-practices.ipynb — new notebook
• doc/memory-best-practices.nblink — nblink entry pointing to the notebook
• doc/index.rst — added memory-best-practices to the User Guide toctree