Add classic structured DAG example (Gaussian elimination, FFT)

[rodrigodg1]

Summary

Adds scripts/examples/classic_dags, an example with generators for the two classic structured task-graph families used to evaluate list schedulers in the heterogeneous-scheduling literature: Gaussian elimination and FFT, with the exact topologies specified in the HEFT/CPoP paper (Topcuoglu, Hariri and Wu, IEEE TPDS 13(3), 2002, doi:10.1109/71.993206, Figs. 8 and 10). The same families appear in the evaluation of PEFT (Arabnejad and Barbosa, IEEE TPDS 25(3), 2014, doi:10.1109/TPDS.2013.57).

• Each generator (gaussian_elimination_dag, fft_dag) asserts the paper's closed-form task count and acyclicity on every run.
• The script schedules each instance with seven schedulers (HEFT, CPoP, PEFT, Min-Min, Max-Min, Sufferage, ETF) and prints a makespan comparison.
• Task-graph and Gantt drawings are sized by instance count so labels stay legible on both small and large graphs (the default drawing parameters overlap badly past ~20 tasks).
• Drawings omit SAGA's internal __super_source__/__super_sink__ (added automatically since the FFT graph has multiple exit tasks) so they match the papers' figures; the dummy tasks are still scheduled and never affect makespans.
• Instances are reproducible via a fixed numpy seed. Two schedulers currently break ties by Python set-iteration order (ETF in general, PEFT on highly symmetric graphs such as FFT), so their rows in the comparison may vary between runs — noted in the example's Readme.
• Adds the corresponding entry to scripts/examples/Readme.md.

Type of change

• Bug fix
• New scheduler
• New feature (other)
• Refactor / maintenance
• Docs / CI only

Checklist

• Branched off main and up to date with it (no conflicts)
• uv run pytest tests/ --timeout=120 passes locally (124 passed)
• uv run mypy src/saga --ignore-missing-imports is clean (unaffected; src/saga is untouched by this PR — also ran mypy directly against the new file, clean)
• uv run ruff check src/saga and uv run ruff format --check src/saga are clean (unaffected; also ran both directly against the new file, clean)
• New or changed behavior is covered by tests
• Public functions/classes are typed and documented in the existing docstring style
• No stray files, debug prints, unintended pyproject.toml/uv.lock changes, or version bump

Notes for reviewers

• Left "covered by tests" unchecked: this is an example script under scripts/, which is outside the pytest/CI-linted src/saga scope (consistent with the other examples in scripts/examples/, none of which have dedicated pytest coverage). The generators self-verify structurally (task-count formula + acyclicity) on every run instead.
• Only the topologies follow the cited papers; task costs and dependency sizes are drawn from SAGA's existing random-weight utilities rather than the papers' cost-assignment protocol (e.g. the HEFT paper assigns level-uniform costs to FFT tasks).
• While preparing this example I noticed ETFScheduler (and PEFTScheduler on symmetric graphs) produce different makespans across runs on identical instances, due to unsorted set iteration in tie-breaking. Happy to open a separate issue/PR for that if useful — didn't want to scope-creep this one.