Remove caching from torch information endpoint

[betatim]

The @cache decorator was not working so we remove it. Because it uses self as part of they key each new instance had a new fresh cache. And basically you get a new instance every time.

Closes #470

[Copilot]

Pull request overview

Removes functools.cache-based memoization from the PyTorch __array_namespace_info__ inspection namespace because it was ineffective in common usage patterns (new __array_namespace_info__ instances led to cache misses).

Changes:

• Drop from functools import cache import.
• Remove @cache decorators from dtypes() and devices() in the PyTorch info namespace.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[ev-br]

LGTM and CI is green and it's merged. Thanks @betatim

[betatim]

Out of interest: where/how is it configured that copilot comes and reviews? And can you configure copilot to tell it things about how to review?

[ev-br]

Yeah, we should prbably just disable these automated Copilot revews
Based on https://github.com/orgs/community/discussions/189857, there's a hierarchy of org level > repo level > user level. For the first two, we need to ask the admins. @kgryte or @rgommers -- could you please disable Copilot reviews for the data-apis org and the array-api-compat repository?

[rgommers]

Disabled at the repo level. There is no org-level setting for Copilot it looks like.