docs: add sparse tensor naming and shape conventions

[theo-barfoot]

Summary

Adds a repository-wide guide for naming sparse tensors, matrix shapes, batch
dimensions, sparse and dense dimensions, layouts, and specified entries.

The guide establishes the following core convention:

• an unbatched sparse matrix has shape (n_rows, n_cols);
• a batched sparse matrix has shape (batch_size, n_rows, n_cols);
• terms such as “2D”, “3D”, “rank 2”, and “rank 3” describe tensor rank only
  and should not be used to imply batching or spatial meaning.

It also documents:

• the distinction between tensor rank, batch dimensions, sparse dimensions,
  and dense value dimensions;
• differences between COO and compressed sparse layout batching;
• nse, nnz, explicit zeros, and structural zeros;
• standard shape notation for multiplication and linear solves;
• reduction-axis terminology;
• coalesced and uncoalesced COO tensors;
• terminology for views, copies, reshaping, and dense materialisation.

Motivation

Recent work exposed some ambiguity between terms such as “3D sparse tensor” and
“batched sparse matrix”. A tensor with shape (batch_size, n_rows, n_cols) is
rank 3, but the leading dimension represents a batch of matrices rather than
a third matrix or spatial dimension.

This guide makes the logical object and its shape explicit, aligns terminology
with existing conventions in the repository, and follows PyTorch sparse tensor
terminology where applicable.

Documentation changes

• adds docs/source/naming.rst;
• links the guide from the main documentation index;
• links the guide from the contributing documentation.

Scope

Documentation only. This PR does not change runtime behaviour or public APIs.

devcontainer

Also changed the devcontainer.json to remove the CPU pinning that is only useful for my personal machine and would be likely to fail for anyone with less than 24 CPU cores/threads.

[Copilot]

Pull request overview

Adds a repository-wide documentation guide that standardizes how sparse tensor/matrix shapes, batching, layouts, and “specified entries” terminology are described across the project, and links it from the main docs and contributing guide.

Changes:

• Add a new naming.rst guide defining shape/batching/layout terminology and recommended wording.
• Link the guide into the Sphinx docs index and the contributing documentation.
• Adjust devcontainer runtime arguments (removes CPU pinning).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
docs/source/naming.rst	New comprehensive naming/shape/layout conventions guide for sparse tensors and batching.
docs/source/index.rst	Adds the new guide to the documentation toctree.
docs/source/contributing.rst	Points contributors to the naming/shape conventions guide.
.devcontainer/devcontainer.json	Removes --cpuset-cpus from container run arguments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.