Add 'in-thread one-line DM precedes sub-agent dispatch on Slava-active PRs' rule to HEARTBEAT.md

Edit ~/agents/slava-agent/HEARTBEAT.md Step 4 (Git & work) to add a new subsection adjacent to the Post-Push-Protocol-all-branches and 3-rounds-then-state-coverage sections. Title: 'In-thread one-line DM precedes sub-agent dispatch on Slava-active PRs'. Content: 'When a Cursor or CodeRabbit finding fires on a Slava-authored PR you are actively monitoring AND Slava is in active-merge mode (he is online, has merged within the past hour, or is in a known active-merge window), the FIRST action is a one-line in-thread DM to him in the existing Slack thread for that PR/topic, the SECOND action is the sub-agent dispatch. DM template (5 seconds): "on it, ~N min, hold merge if you can" where N is your gate+push estimate (typically 3-8 minutes for a single-finding fix). Why: if you do the sub-agent dispatch silently, Slava sees the same Cursor notification independently and may merge before your fix lands. The sub-agent gate-and-push window is typically 3-8 minutes; Slava merge-on-clean-head latency is typically 1-3 minutes from a notification. Without the in-thread one-liner, you are racing his attention. Case study (2026-06-08 PR #11886, BOLT-1211): Cursor finding posted on a Slava-authored PR he was actively merging. Sub-agent was dispatched silently. Local gates took ~6 minutes. Slava merged ~30 seconds before the fix push could land on BOLT-1211/adapters-oem. Recovery: cherry-pick the orphan onto cato/bolt-1211-staleness-on-failed-polls, open follow-up PR #11888 (all CI green, REVIEW_REQUIRED, 14h post-open at reflection time). The one-line DM (5 seconds) would have prevented the orphan PR. When NOT to do this: if the PR is dormant (no merge activity in past several hours, Slava offline / out-of-hours), or if the finding is a nitpick (CodeRabbit Low), the silent sub-agent dispatch is fine. The rule is specifically for Slava-active + Medium+ findings.' Delivery: workspace-internal write to ~/agents/slava-agent/HEARTBEAT.md. No PR. After the edit, PATCH this fleet-task to completed using bin/fleet-task-patch.sh with a short result citing the case study.