fix: replace deprecated ImageAnalysisToolkit with ScreenshotToolkit i…

[Wendong-Fan]

…n multi-modal

The multi-modal agent's image analysis was silently using a hardcoded gpt-4.1-mini model instead of the
user-configured model. This happened because ImageAnalysisToolkit creates its own internal model instance,
falling back to CAMEL's default when none is provided.

The fix follows the same approach already used by developer, browser, and document agents (introduced in #1196):
replace ImageAnalysisToolkit with ScreenshotToolkit, which delegates image understanding to the main agent's
model via toolkits_to_register_agent. This avoids the separate model instantiation entirely.

Also removes all references to the now-unused ImageAnalysisToolkit across backend, server config, and frontend.

Changes

• Replace ImageAnalysisToolkit with ScreenshotToolkit in multi-modal agent factory
• Delete image_analysis_toolkit.py and clean up all imports/references
• Update multi-modal system prompt with correct tool names (read_image, take_screenshot_and_read_image)
• Update frontend workflow config, icon mappings, and test mocks
• Remove IMAGE_ANALYSIS from server-side config group