Remove GPU dynamic shapes restriction

[helena-intel]

• Remove message about GPU dynamic shapes. Dynamic shapes on GPU work in the current OpenVINO prerelease, though there are still some known issues. We currently only show this message when a user loads a model though, not when moving a model .to("gpu").
• Change comment for .half(). It's not necessary anymore to explicitly convert the model to FP16 for increased performance on GPU; OpenVINO will do this automatically. It's still a useful method to save the model weights with half the size.

We should discuss how to support setting explicit inference hints https://docs.openvino.ai/latest/openvino_docs_OV_UG_supported_plugins_CPU.html

[echarlaix]

Thanks a lot for updating it @helena-intel, could you integrate it in the documentation as well ?

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.