feat: add Avian as LLM service provider

[avianion]

Summary

• Adds AvianService as a new LLM service option for the Avian inference API (https://api.avian.io/v1), which is OpenAI-compatible
• Follows the same patterns as the existing OpenAIService and AzureOpenAIService implementations
• Adds corresponding test in tests/services/test_service_init.py

Available Models

Model	Context Window
DeepSeek V3.2 (deepseek-v3.2)	164K
Kimi K2.5 (kimi-k2.5)	128K
GLM-5 (glm-5)	128K
MiniMax M2.5 (minimax-m2.5)	1M

Usage

marker_single input.pdf \
  --llm_service marker.services.avian.AvianService \
  --avian_api_key YOUR_API_KEY \
  --avian_model deepseek-v3.2

Configuration

Option	Description	Default
avian_api_key	API key for the Avian service	(required)
avian_model	Model to use	deepseek-v3.2
avian_image_format	Image format for multimodal requests	png

Test plan

• Added test_llm_avian to tests/services/test_service_init.py verifying service initialization
• Verify structured output (JSON schema) works with Avian's OpenAI-compatible endpoint
• Verify image processing works for multimodal PDF pages

[github-actions]

CLA Assistant Lite bot All contributors have signed the CLA ✍️ ✅

[avianion]

I have read the CLA Document and I hereby sign the CLA

[avianion]

Addressed feedback: corrected minimax-m2.5 token limits in docstring — changed from "1M context" to "200K context, 16K max output" to accurately reflect MiniMax M2.5 (MiniMax-Text-01) capabilities.