add aifunction backprop demos on DS-1000 examples

[shuaichenchang]

Summary

• Add two DS-1000 backpropagation demos showing the full ai_function learning loop
  (empty-memory baseline → 8-example training → re-test with trained memory):
  • examples/scipy_backprop_demo.py : an example in scipy domain
  • examples/pandas_backprop_demo.py: an example in pandas domain
  • examples/ds1000_utils.py: a shared helper wrapping the DS-1000 executor, parameter recall, parallel batch runner, and optimizer feedback builder.
• Add scipy>=1.14.0 and scikit-learn>=1.5.0 to examples/pyproject.toml
• Update README.md and docs/tutorial.md with short desciptions of new examples

Test Plan

• uv sync --all-packages resolves scipy and scikit-learn
• uv run examples/scipy_backprop_demo.py — scipy_787 goes FAIL → PASS
• uv run examples/pandas_backprop_demo.py — pandas_263 goes FAIL → PASS

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.