feat: add OOM pre-check for vision models and fix InternVL image dime…

[sufubao]

…nsion handling

[gemini-code-assist]

Code Review

This pull request introduces an OOM (Out of Memory) pre-check mechanism for Qwen-family vision models by performing a dummy forward pass with worst-case image dimensions during initialization. It updates several model implementations to support this check and refactors the ViT model to derive inference parameters from the configuration instead of environment variables. Feedback focuses on improving memory management within the pre-check function by explicitly deleting tensors and clearing the CUDA cache, as well as refining exception handling and logging practices.