ai_provider_llama_cpp 1.2.x-dev

*First beta of the 1.2 series**

This is a **beta** release. It is recommended to test it in development or staging environments before using it in production.

### Major changes

- **Multi-server (multi-instance) support**:
You can now configure multiple llama.cpp / OpenAI-compatible servers. Each server is exposed as an independent provider (e.g. `llama_cpp:gpu`, `llama_cpp:embeddings`).
- Important breaking change: The plugin ID changed from `llama_cpp` to `llama_cpp:{server_id}`.
- Includes an update hook (10201) that automatically migrates legacy configuration.

- **Key module integration** (update 10202):
API keys are no longer stored as plaintext. They are now managed as secure Key entities.

- **New capabilities**:
- Text to Image generation
- Rerank
- Moderation with native support for LlamaGuard3 and ShieldGemma (including multiple safety policies)

- **Per-server model management**:
- Model filtering using patterns
- Manual operation type overrides per model
- Improved automatic capability detection

- **Stability and improvements**:
- Better derivative handling and per-server caching
- Configurable timeout and authentication per server
- Type safety and static analysis improvements (PHPStan)
- Documentation updates

### Requirements

- Drupal 10.2, 11 or 12
- AI module ^1.2
- Key module ^1.18

### Feedback

Please test this beta and report any issues in the project's issue queue.

A stable 1.2.0 release is planned once this beta has been validated.