Update tool result handling to support rich content types (images, PDFs)

[gadenbuie]

Background

The ellmer package recently added support for tool results that return non-JSON content types like images and PDFs (ellmer PR #735).

Problem

Currently, shinychat tries to convert tool results to JSON strings using tool_string(), which fails when the tool result contains S7 content types like content_image() or content_pdf().

Proposed Solution

Modify the tool result handling to:

• Detect S7 content types
• Render these content types appropriately in the chat interface
• Fallback to existing JSON conversion for other content types

Example Scenario

library(ellmer)

# Tool that returns an image
cat_image_tool <- tool(
  function() {
    # Simulate generating a cat image
    tmpf <- tempfile(fileext = ".jpg")
    download.file("[https://placekitten.com/200/300](vscode-file://vscode-app/Applications/Positron.app/Contents/Resources/app/out/vs/code/electron-browser/workbench/workbench.html#)", tmpf, quiet = 
TRUE)
    content_image_file(tmpf)
  },
  name = "get_cat_image",
  description = "Generate a random cat image"
)

# Create a chat
chat <- chat_openai()
chat$register_tool(cat_image_tool)

shinychat::chat_app(chat)

Related Work

• ellmer PR: feat: Support images and PDFs in tool results tidyverse/ellmer#735
• Specific issue: Handling of tool_string() for S7 content types

/cc @gadenbuie

[cpsievert]

Chatlas also recently added support posit-dev/chatlas#231