Update AI-hosting model catalog: add GLM/Qwen/Mistral entries with corrected IDs and capabilities, remove Devstral-small-2-24B

[Copilot]

mittwald AI Hosting introduced new models and retired Devstral-small-2-24B. This PR updates the provider’s hardcoded model capability mapping and docs so exposed models align with the current catalog.

• Model capability mapping (includes/MittwaldModelMetadataDirectory.php)

  • Removed retired model mapping:
    • Devstral-Small-2-24B-Instruct-2512
  • Added/updated model mappings:
    • text/chat options: Qwen3.5-0.8B
    • multimodal options: Mistral-Medium-3.5-128B
    • reranking model (no chat capabilities/options): Qwen3-VL-Reranker
    • OCR model with dedicated restricted options/capabilities: GLM-OCR

• Public model docs (README.md, readme.txt)

  • Replaced Devstral mention with the new model set.
  • Corrected explicit model IDs to:
    • Mistral-Medium-3.5-128B
    • Qwen3.5-0.8B
    • GLM-OCR

case 'Qwen3.5-0.8B':
	$modelCaps    = $gptCapabilities;
	$modelOptions = $gptOptions;
	break;

case 'Mistral-Medium-3.5-128B':
	$modelCaps    = $gptCapabilities;
	$modelOptions = $gptMultimodalInputOptions;
	break;

case 'GLM-OCR':
	$modelCaps    = $gptOcrCapabilities;
	$modelOptions = $gptOcrOptions;
	break;

case 'Qwen3-VL-Reranker':
	$modelCaps    = array();
	$modelOptions = array();
	break;

[LukasFritzeDev]

@copilot fix the open conversations

[martin-helmich]

@copilot Fix the issues from code review. Read the documentation FIRST. Make sure to use the exact model IDs as documented in the documentation. Also, pay attention to the documented model capabilities.

[Copilot]

@copilot Fix the issues from code review. Read the documentation FIRST. Make sure to use the exact model IDs as documented in the do...

Done in 596abc8: model IDs now use Qwen3.5-0.8B and Mistral-Medium-3.5-128B, Mistral Medium is mapped with multimodal capabilities, and GLM-OCR is handled separately with its own capability/options mapping.

@copilot fix the open conversations

Addressed in 596abc8 with the open review fixes in model mapping and docs.