Feature Request: Add child spans for detailed OTEL traces

Problem

Currently, OpenClaw's OTEL instrumentation only provides coarse-grained spans. For example, openclaw.message.processed is a single span that encompasses the entire message processing time (~228 seconds), but there is no breakdown of where that time is spent.

Current Spans Observed

• openclaw.message.processed - entire message processing (no child spans)
• openclaw.model.usage - model API call with attributes but no sub-steps
• openclaw.session.stuck - session stuck detection

Desired Behavior

Add child spans under openclaw.message.processed to break down:

1. Tool calls - time spent in each tool execution
2. Model API latency - time for API request/response round-trip
3. Tokenization - time spent calculating/counting tokens
4. Response building - time spent constructing the final response
5. Other sub-operations - any significant internal steps

Example Use Case

When debugging slow responses, developers need to understand where time is spent:

• Is it waiting on an LLM API?
• Is it running tool executions?
• Is it processing tokens?

Proposed Implementation

Wrap significant internal operations with child spans:

const parentSpan = tracer.startSpan("openclaw.message.processed");
// ...
const toolSpan = tracer.startSpan("openclaw.tool.execution", { parent: parentSpan });
// tool work
toolSpan.end();
// ...
parentSpan.end();

Environment

• OpenClaw Version: 2026.4.5
• OTLP Backend: Jaeger
• Protocol: http/protobuf