[KAVentures]

The default compact hosted workflow-evidence response omitted PR #260's map, and a tail of capture gaps could hide earlier useful work from orientation. Large selected histories also lacked a bounded analysis plan and visible loading progress.

This change:

• Includes the existing bounded agent_workflow_map in compact responses, preserving its authority and uncertainty boundaries.
• Preserves existing recent rows and adds separate latest_work and capture_health sections. The work query is bounded, actor-scoped and retention-filtered; gaps do not imply recorder liveness or work time.
• Adds date chunks intersected with the selected handoff window, a 31-date listing limit with an explicit continuation flag, and instructions for bounded raw pagination and coverage tracking.
• Shows indeterminate loading progress and cancels obsolete preview requests when the selected period changes. Full-period downloads and existing history permissions remain unchanged.

Validation:

• Full combined suite: 1,884 passed, 1 skipped. Final targeted map, gap-tail, privacy and export checks also passed.
• Authenticated MCP replay: 48 synthetic cases, complete raw-event retrieval, all 16 tested maps returned in compact mode, and earlier work recovered in all 8 gap-tail cases.
• Chromium check: loading state, period switching, chunk guidance and permission text passed with no JavaScript errors.
• Authenticated preview function: 300,000-event synthetic SQLite history, 31 date chunks plus an explicit extra-dates flag, approximately 1.3 seconds on the test VM. This is not an OS capture or production latency benchmark.

This PR depends on #261; merge that privacy PR first. No user evidence is committed. Grouping remains heuristic, and these checks do not establish AI task-completion uplift or measured savings.