Support extracting non-text content from chat responses

[thisisnic]

When an LLM returns content like images and audio from chat responses, there's no ellmer API for extracting and saving it, though it's possible. Example workarounds:

• Example code to generate WAV audio from text using Gemini TTS and present it in a Shiny audio preview modal #937
• https://github.com/hadley/bananarama/blob/main/R/bananarama.R#L246