feat(trace-aliyun): GenAI spans for LLM TTFT, agent ReAct, tools, and failed LLM nodes

[123liuziming]

Important

1. Make sure you have read our contribution guidelines
2. Ensure there is an associated issue and you have been assigned to it
3. Use the correct syntax to link this PR: Fixes #<issue number>.

Summary

Improve Aliyun Trace (trace-aliyun) so workflow/agent telemetry better matches GenAI semantic conventions and is more useful when model or tool calls fail.

LLM

• Record gen_ai.response.time_to_first_token (nanoseconds) on workflow LLM spans and message LLM spans when TTFT is available.
• On LLM invoke failure (when graphon leaves process_data empty), fall back to node inputs / error / error_type so attributes such as gen_ai.request.model, gen_ai.provider.name, gen_ai.prompt, and gen_ai.completion are still populated when possible.
  • Note: structured gen_ai.input.messages still depends on process_data.prompts, which is only written after a successful invoke today.

Agent / ReAct

• Emit AGENT spans (instead of TASK) for agent-strategy nodes, with gen_ai.operation.name=invoke_agent, agent name, and token usage.
• Parse agent execution logs in node outputs to build per-round ReAct STEP spans and child call spans, anchoring monotonic log timestamps to the node start time.
• Classify agent log entries correctly:
  • Thought → LLM
  • CALL … → TOOL (previously misclassified as LLM because CALL logs also set metadata.provider)

Tool

• Align tool span attributes with GenAI conventions:
  • Required: gen_ai.span.kind=TOOL, gen_ai.operation.name=execute_tool
  • Recommended/optional: gen_ai.tool.name, gen_ai.tool.type, gen_ai.tool.description, gen_ai.tool.call.id, gen_ai.tool.call.arguments, gen_ai.tool.call.result
• Apply across message tool_trace, workflow tool nodes, and agent ReAct CALL spans.
• Remove legacy tool.name / tool.description / tool.parameters keys.

Tests

• Unit coverage for TTFT, agent STEP/LLM/TOOL span building, failed-LLM fallback, GenAI tool attribute helpers, and semconv constants.

Screenshots

Before	After
Agent CALL logs shown as LLM; failed LLM spans missing model/prompt; tool attrs used legacy tool.* keys	Agent CALL → TOOL; failed LLM spans keep model/provider/inputs when available; tool attrs use gen_ai.tool.*

Checklist

• This change requires a documentation update, included: Dify Document
• I understand that this PR may be closed in case there was no previous discussion or issues. (This doesn't apply to typos!)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.
• I ran make lint && make type-check (backend) and cd web && pnpm exec vp staged (frontend) to appease the lint gods

From Cursor

[123liuziming]

@TreyDong @laipz8200 Hey, could you review my PR when you have a chance?