Added batch predict for Imagesummarizer

[mogith-pn]

Description

• ImageSummarizer tool is failing for large PDFs or PDFs with more images, since it is doing single batch predict.
• Changed the Imagesummarizer class function to have batch predicts.
• Added a filter identifier in order to filter queries while doing multimodal search.
• For ex
  res = search.query(filters=[{'input_types': ['text'], 'input_dataset_ids':['d3'], "metadata": {"image_summary": "yes"}}], ranks=[{"text_raw": qstn}])
  In above search query we filter the summary text from normal document chunk by the metadata, which was added in the ImageSummarizer pipeline.

https://clarifai.atlassian.net/browse/DEVX-870

[sanjaychelliah]

👍