Support truncating when the input size exceeds model limit

[pengjiang80]

Some Application frameworks can't handle special characters very well and may not follow the chunk size strictly. In some cases, the size of embedding or reranking may exceed the setting and cause error based on model configuration. As it's a bug at the application level and may or may not be fixed, we can't guarantee that using a model with a bigger context size and setting a bigger ubatch size can 100% resolve the problem. Some users are willing to bear some accuracy loss and embed the document without error. We can add a parameter to handle this as both ollama and some other platforms support this. It's a setting not recommended, but users can turn it on at their own risk.

Refer to gpustack/gpustack#950