Surface Errors on Tool/LLM Spans in VercelAI SDK Integration

[matejminar]

Description

We've noticed in the new Agent Monitoring that if there's an error thrown inside the execute tool span it's not visible in the UI.
We'd like to display some indicator on span itself (fire icon) + ideally link to sentry issue.

We also need a way to plot these errored spans on a chart so something like span.status:error would be helpful because it enables both usecases. But open to suggestions here.

Same goes for llm calls (e.g generate_text span). We need to validate if provider returns an error (like wrong model, quota exceeded, rate limited, etc.) this needs to be similarly surfaced on the span so that we can take action in UI.