Add a guide for creating new conversion packages for rosidl::Buffer (backport #6654)

[mergify]

Description

Add a new page for a guide to creating a new conversion packages for rosidl::Buffer.

The page explains how conversions packages differ from rosidl::BufferBackend plugins: backends handle storage and transport, while conversions packages adapt existing ROS messages with uint8[] payload fields, such as tensors, images, or point clouds, into library-specific native types.

It also includes tensor-specific guidance for tensor_msgs/msg/ExperimentalTensor, documents torch_conversions as the first reference implementation, and outlines how future packages such as ONNX, NumPy, or CuPy conversions can interoperate through the common tensor message while using CPU or supported non-CPU buffer backends underneath.

Did you use Generative AI?

Yes, GPT-5.5 was used to help create an initial draft of the content.

---

This is an automatic backport of pull request #6654 done by [Mergify](https://mergify.com).

[github-actions]

HTML artifacts: https://github.com/ros2/ros2_documentation/actions/runs/25678707880/artifacts/6922277505.

To view the resulting site:

1. Click on the above link to download the artifacts archive
2. Extract it
3. Open html-artifacts-6682/index.html in your favorite browser