feat: add structured output tutorial (#14)

[builder-ujaladi]

Summary

Adds a new tutorial 14-structured-output to python/01-learn/ covering structured output with Strands Agents.

What's covered

1. Flat Pydantic models — define a model, pass as structured_output_model, access result.structured_output (sync + async)
2. Complex schemas — nested models, List[SubModel], Optional fields, Field validators, Enum/Literal constraints
3. Validation & self-correction — automatic retry loop on validation failure, StructuredOutputException handling, custom structured_output_prompt
4. Tools + structured output — calculator tool from strands-agents-tools combined with structured results
5. Streaming — stream_async() with structured output (appears only in final event)

Files added

• python/01-learn/14-structured-output/README.md
• python/01-learn/14-structured-output/structured-output.ipynb
• python/01-learn/14-structured-output/requirements.txt
• python/01-learn/14-structured-output/images/architecture.png

Testing

• Notebook executed end-to-end with nbconvert --execute in a clean venv — all 18 code cells pass with zero errors
• Manual testing also completed
• Model: Claude Sonnet 4.5 on Amazon Bedrock (us.anthropic.claude-sonnet-4-5-20250929-v1:0)

[github-actions]

Latest scan for commit: 6672e91 | Updated: 2026-04-19 18:37:38 UTC

✅ Security Scan Report (PR Files Only)

Scanned Files

• python/01-learn/14-structured-output/README.md
• python/01-learn/14-structured-output/images/architecture.png
• python/01-learn/14-structured-output/requirements.txt
• python/01-learn/14-structured-output/structured-output.ipynb

---

Security Scan Results

Critical	High	Medium	Low	Info
0	0	0	0	0

Threshold: High

No security issues detected in your changes. Great job!

This scan only covers files changed in this PR.

[manoj-selvakumar5]

The architecture diagram tries to show everything at once, which feels a bit overwhelming. Could you simplify it to represent just the core concept? You pass a model, you get a typed object back — that's the "aha moment."

[manoj-selvakumar5]

• I'm wondering if the streaming section make sense as a standalone section since there is no visible streaming happening? Would it make sense to just add as a note, by combining streaming with for example previous section. The section is now teaching "structured output doesn't stream."

• Across the tutorial, the code cells print results but doesn't explain what the developer should take away from the output. For example, in the constraints section, the output shows Confidence: 95% but without a callout, there is no way to know immediately that the raw value is actually 0.95 displayed through :.0% formatting. The same pattern repeats in the self-correction and other sections: the output appears, and the notebook moves on. A short markdown cell after each output connecting what the developer sees back to the concept being taught would make a big difference. Something like: "Notice that X happened because of Y" "Notice that sentiment is exactly one of the three allowed Literal values and not somewhat negative or frustrated".

• The self-correction section claims the LLM is "likely to violate" the validator on its first attempt, but all three cells show Tool #1 only the LLM gets it right every time. The field_validator never actually rejects anything. Could you please try it a bit differently so the LLM guess and fail? probably add a cell that prints agent.messages after the call to show the retry trail, use a harder constraint the LLM is genuinely likely to violate??