feat: add DeepSeek V3.2 and MiniMax M2.1 models to NVIDIA provider

[starfield17]

Closes #8647 Added two new models to the NVIDIA provider:

• nvidia/deepseek-ai/deepseek-v3.2 - DeepSeek V3.2 with reasoning and tool call support
• nvidia/minimaxai/minimax-m2.1 - MiniMax M2.1 with advanced reasoning capabilities Changes
• Modified packages/opencode/test/tool/fixtures/models-api.json
• Added complete model configurations with cost, context limits, and capabilities
• JSON format validated successfully Models Details DeepSeek V3.2
• Context: 163,840 tokens
• Output: 65,536 tokens
• Cost: $0.28 input / $0.4 output
• Features: reasoning, tool_call, temperature
• Open Weights: true
• Family: deepseek-v3 MiniMax M2.1
• Context: 204,800 tokens
• Output: 131,072 tokens
• Cost: $0.3 input / $1.2 output
• Features: reasoning, tool_call, temperature, interleaved reasoning
• Open Weights: true
• Family: minimax
• Special: interleaved: { field: "reasoning_details" } Testing
• JSON syntax validated ✓
• Model configurations match existing patterns ✓
• All required fields included ✓ This PR adds comprehensive support for these two new models within the NVIDIA provider ecosystem.