whaleflow: TUI /workflows monitoring surfaces — run/inspect/replay

[Hmbown]

Pod-style background workflow monitoring in the TUI: list runs with done/total progress, longest-running item peek, inspect a run, trigger replay/report.

Acceptance: a running workflow is observable and replayable from the TUI behind the feature flag.

Part of the v0.9.0 EPIC.

[Hmbown]

Kimi Code audit note, 2026-06-12:

The TUI monitoring surface should treat a workflow/run as the first-class object and child agents as projected headless lanes. Kimi's subagents are mostly isolated child runs with persisted trace/output and lifecycle events; the UI projects that state instead of making the child UI card the canonical state.

For /workflows inspect/replay, I would explicitly require:

• run-level summary first, lane drilldown second;
• per-lane trace/evidence handles, not just clipped preview text;
• live compact progress while lanes run, so the parent/user do not see opaque Running loops;
• clear distinction between child self-report, verified evidence, and UI-truncated output;
• replay/export anchored to workflow_run_id + node id + agent/session id.

Related Kimi-informed slice: #3180.