Add Blip2 model in VQA pipeline (#25532)
* Add Blip2 model in VQA pipeline
* use require_torch_gpu for test_large_model_pt_blip2
* use can_generate in vqa pipeline
* test Blip2ForConditionalGeneration using float16
* remove custom can_generate from Blip2ForConditionalGeneration
