feat(otel): stamp tangle.run.id + tangle.subject.key on the loop root span

[drewstone]

What

Two additive attributes on the loop root span in otel-export.ts:

• tangle.run.id — explicit run identity (= runId), so a consuming system of record maps a span tree to exactly one run deterministically, rather than inferring it from gen_ai.conversation.id.
• tangle.subject.key — the subject grain (= agentRunNames[0], the lead agent/profile the run drove), so runs group under the thing being worked on instead of one service-wide bucket.

Why

Tangle Intelligence's new agent-work spine reconstructs subjects + runs from ingested OTel spans. Today it can group agent-runtime runs by conversation id and derive a subject from service.name — but that buckets every run under one subject. Stamping the run id + subject explicitly lets the spine map run → subject from the executor itself, deterministically.

Safety

• Purely additive — both values come from data already in the loop.started payload (runId, agentRunNames). No new plumbing, no behavior change.
• Existing attributes untouched; the 17 otel-export tests pass, with two added assertions for the new keys.

Pairs with the Intelligence spine PR (tangle-network/agent-dev-container#2504). This is the optional "deterministic mapping" half — the spine already works without it by inference; this makes the mapping exact.

[tangletools]

✅ Auto-approved PR — 750b69c5

Blanket team auto-approval is enabled for this reviewer service.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: blanket_auto_approve · 2026-06-22T00:13:28Z}