Add Amazon Bedrock Mantle OpenAI Responses API sample

## Summary

Add an introductory, end-to-end sample showing how to use the OpenAI Python SDK with the Amazon Bedrock `bedrock-mantle` endpoint and Responses API.

## Motivation

Amazon Bedrock supports OpenAI-compatible Models and Responses APIs through the regional Mantle endpoint. The repository currently documents the native InvokeModel and Converse APIs, but does not include a focused introductory notebook for migrating an OpenAI SDK workflow to the Bedrock endpoint.

## Proposed scope

- Add a self-contained notebook under `introduction-to-bedrock`.
- Add the corresponding website Markdown page required by the contribution guidelines.
- Demonstrate regional endpoint configuration and secure API-key handling.
- Discover and validate available model IDs.
- Cover non-retained requests, stored multi-turn state, response retrieval, streaming, and background processing.
- Include bounded polling, actionable error handling, security guidance, cost awareness, and cleanup.
- Add the sample to the introductory README and website navigation.

## Acceptance criteria

- The notebook and website page contain all required contribution-guide sections.
- The notebook contains no credentials, saved outputs, or execution counts.
- Examples use a documented Responses API-compatible model and explicitly constrain output tokens.
- The documented minimum OpenAI SDK version supports every parameter used by the sample.
- Notebook syntax, documentation rendering, internal navigation, and external links validate locally.

## Operational notes

The sample will not provision infrastructure. Running inference requires a regional Amazon Bedrock API key and can incur model usage charges.

Begin by reviewing the existing notebooks under introduction-to-bedrock, the contribution guidelines, the introductory README, and website navigation. Run the local notebook syntax, documentation rendering, internal navigation, and external-link validations. Done means the notebook and website page meet all acceptance criteria without credentials, saved outputs, or execution counts.

Written by the indexing model from the issue text.