P+: Extended Textual Conditioning in Text-to-Image Generation