Fine tuning support

[dillonalaird]

A couple features and fixes:

• Keeps fine tune florence2 in meta tools and enables the florence2 tool function to call the fine tuned endpoint
• Allows the orchestrator to edit code to utilize the new fine tuned model
• Fixes bug in local code interpreter to ensure the execution path is the remote_path
• Updates prompts to ensure model prints function output to view results (previous it would expect to see results if it returned them, if it didn't see it this would lead to a long debug loop)
• Fixes bug where run_code was not returning errors to agent
• Fixes bug in edit_artifact so it properly outputs error messages for bad edits