docs: PyTorch Compute Platform Quality Levels

As previosly discussed at:
* TAC meeting 10/14/2025, https://lists.pytorch.org/g/tac/topic/agenda_for_october_14_2025/115753112
* Accelerator Integration Workking Group meeting 11/5/2026
* pytorch/rfcs#63 RFC

As a leading deep learning framework PyTorch capabilities are regularly expanded with
the support for new compute platforms and device backends. This commit introduces scoring
criteria to assess quality levels of PyTorch device backends to help developers identify
gaps and equip them with the tool to make decisions whether certain compute platforms are
ready for specific milestones.

Scoring covers requirements for platform hardware and software availability, maturity,
support obligations, platform features, ci coverage, etc.

Compute Platforms quality levels are defines as follows:

* **Engineering** compute platforms
* **Unstable** compute platforms
* **Stable** compute platforms

**Engineering** platforms are platforms which are under active development and might not be ready for adoption. **Unstable** and **Stable** platforms are more mature platforms which are ready for adoption with different levels of maturity.

The following table provides requirements for each quality level:

* **Stable** platforms:
  * Satisfy all 100% P0 requirements
  * Reach 80% overall score
* **Unstable** platforms:
  * Reach 70% overall score
* **Engineering** platforms:
  * Less than 70% overeall score

Signed-off-by: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>

Author: dvrogozh