docs: use scipy optimizer in QAOA MaxCut example

[spital]

Summary

• replace the random QAOA parameter sampling loop with a scipy.optimize.minimize variational loop using COBYLA
• split QAOA execution from the optimizer objective so eval_qaoa_energy returns only scalar energy
• run one final QAOA pass with the optimized parameters for validation and plotting
• document the classical optimizer vs. Guppy quantum-program boundary, including the angle coefficient convention
• add SciPy to the examples dependency group and refresh uv.lock

The performance/recompilation limitation remains documented in the notebook and is intentionally left to #1546.

Testing

• uv lock --check
• uv run ruff check
• uv run ruff format --check
• uv run pytest tests/integration/test_notebooks.py
• uv run jupyter nbconvert --to notebook --execute --ExecutePreprocessor.timeout=1200 --output-dir /tmp/tmp.P7p9GROppQ --output qaoa_executed.ipynb examples/qaoa_maxcut_example.ipynb

Manual QAOA notebook execution reproduced the committed output:

array([5.099, 1.736, 2.63 , 4.751, 4.831, 2.361, 2.2  ])
'Success ratio: 0.4015'

AI disclosure

I used Codex, Claude, and Hermes as AI assistants to help prototype, review, compare related GitHub state, and draft this PR text. I manually reviewed the resulting diff, checked the GitHub issue/PR state, and locally ran the tests and notebook execution listed above. The implementation and PR are submitted with human oversight; any remaining mistakes are mine.

Closes #1545