[tokenizer] Ensure GPU is used in TextEmbeddingTranslator

[david-sitsky]

This fixes #3211, where using djl://ai.djl.huggingface.pytorch/sentence-transformers/clip-ViT-B-32-multilingual-v1 on 0.28.0 fails on a GPU machine with:

Caused by: ai.djl.engine.EngineException: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_mm)
	at ai.djl.pytorch.jni.PyTorchLibrary.torchNNLinear(PyTorchLibrary.java) ~[pytorch-engine-0.28.0.jar:?]
	at ai.djl.pytorch.jni.JniUtils.linear(JniUtils.java:1376) ~[pytorch-engine-0.28.0.jar:?]
	at ai.djl.pytorch.engine.PtNDArrayEx.linear(PtNDArrayEx.java:397) ~[pytorch-engine-0.28.0.jar:?]
	at ai.djl.huggingface.translator.TextEmbeddingTranslator.processEmbedding(TextEmbeddingTranslator.java:181) ~[tokenizers-0.28.0.jar:?]
	at ai.djl.huggingface.translator.TextEmbeddingTranslator.batchProcessOutput(TextEmbeddingTranslator.java:144) ~[tokenizers-0.28.0.jar:?]
	at ai.djl.inference.Predictor.batchPredict(Predictor.java:190) ~[api-0.28.0.jar:?]
	at ai.djl.inference.Predictor.predict(Predictor.java:132) ~[api-0.28.0.jar:?]