cendor-sdk 1.10.0

Richer run spans for the monitor: agent-named, step-numbered call spans, a live tree at parity with the post-hoc one, and an opt-in run label. Backward-compatible (additive). Dependency floors bumped to the V2 emission wave (cendor-tokenguard >= 1.3, cendor-acttrace >= 1.7, cendor-guardrails >= 1.6).

Added

• Run label (live_spans(label=…) / span_tree(result, label=…)) → cendor.run.label on the root agent.run span, so a monitor can show what a run was for. It is a deliberate, short, non-sensitive string you choose — never derived from the prompt (prompts/tool values stay off spans).
• Agent name + step index on every call span. chat/execute_tool spans now carry gen_ai.agent.name (which agent made the call) and a 1-based cendor.step ordinal, in both span_tree and live_spans.
• live_spans parity with span_tree. The live root now carries cendor.run.id / cendor.trace_id (learned from the run's correlation id) and, at close, the run's gen_ai.usage.input_tokens / output_tokens and cendor.run.cost_usd rollups.

Changed

• The per-agent max_usd ceiling opened by the SDK is now named (agent:<name> max_usd), so a block by an agent's cap is attributable in a monitor (via tokenguard 1.3's budget(name=)).