feat(providers): add Latitude.sh as new provider

[gsalberto]

Summary

• Add Latitude.sh (lsh.ai) as a new inference provider with OpenAI-compatible API endpoints
• Latitude offers bare metal GPU infrastructure (NVIDIA RTX 6000 PRO Blackwell) with vLLM
• 7 models included across Qwen, Llama, Gemma, and DeepSeek families

Models Added

Model	Family	Input	Output	Context	Features
Qwen 2.5 7B	qwen	$0.27	$0.27	131K	tool_call
Llama 3.1 8B	llama	$0.16	$0.16	128K	tool_call
Qwen3 32B	qwen	$0.45	$0.90	131K	tool_call
Gemma 2 27B	gemma	$0.45	$0.45	8K	tool_call
DeepSeek R1 Distill 14B	deepseek-thinking	$1.44	$1.44	64K	reasoning
Qwen 2.5 Coder 32B	qwen	$0.72	$0.72	131K	tool_call
Qwen 2.5 VL 7B	qwen	$0.36	$0.36	32K	vision

Test plan

• Ran bun ./packages/core/script/validate.ts - all validations pass
• Verified API endpoint returns proper OpenAI-compatible error format
• Logo SVG follows fill="currentColor" pattern