feat: add AkashML provider and models

[fenilmodi00]

feat: add AkashML managed inference provider

AkashML is a managed AI inference service built on the Akash Network decentralized GPU supercloud. This PR adds the provider configuration, a refined logo, and five core models.

What's included

1. Provider Core (providers/akashml/provider.toml)

• Uses @ai-sdk/openai-compatible for drop-in integration.
• API base: https://api.akashml.com/v1.
• Authentication via AKASHML_API_KEY.

2. Refined Logo (providers/akashml/logo.svg)

• Extracted graphical mark from official branding (akashml_logo_long.svg).
• Stripped text for a clean icon-only presentation.
• Optimized with currentColor for seamless light/dark theme support.

3. Model Definitions (5 models)

Added models in vendor-specific subdirectories with verified April 2026 specifications:

Model ID	Family	Context	Modalities	Attributes
meta-llama/Llama-3.3-70B-Instruct	llama	131k	Text	tool_call
deepseek-ai/DeepSeek-V3.2	deepseek	131k	Text	reasoning, tool_call
google/gemma-4-31b-it	gemma	131k	Text, Image, Video	reasoning, attachment, tool_call
qwen/qwen3.5-35b-a3b	qwen	131k	Text, Image, Video	reasoning, attachment, tool_call
minimax/minimax-m2-5	minimax	131k	Text	reasoning, tool_call

Key Notes

• Capabilities Verified: Re-verified reasoning ("thinking" mode) and multimodal support for the 2026 frontier models.
• Pricing: Initial competitive pricing sourced from AkashML; subject to upstream updates.
• Organization: Followed repository best practices for nested model folders.

References

• AkashML Documentation
• Akash Network Blog