Make reporting handle large datasets (payload scale, pagination, streaming)

[vybe]

Summary

Make the reporting pipeline hold up when a report carries a large dataset. Today the payload is a single JSON blob capped at 256 KB (413 over cap) stored in one TEXT column — fine for KPIs, but it fails for a report of thousands of rows.

Context

Part of the Agent Reports epic. As agents report real tabular results (lead lists, scan results, transaction histories), the 256 KB cap and single-blob storage/transfer become the ceiling. This issue covers scaling the create, store, list, and render paths — without breaking the small-report common case.

Acceptance Criteria

• Reports well beyond today's 256 KB cap are supported (raise/stream the cap, or chunk/paginate the payload) with a defined new ceiling
• List endpoints stay metadata-only (no payload) so the Reports tab is unaffected by large payloads (preserve the ReportSummary vs full-payload split)
• Detail/render path paginates or lazy-loads large tabular payloads instead of shipping the whole blob at once
• Create path validates/streams large bodies without pinning a backend coroutine or blowing memory
• Retention sweep still handles large rows efficiently (chunked delete already exists)
• Backward compatible with existing small reports

Technical Notes

db/reports.py storage model (consider off-row / paginated storage for tabular payloads), services/report_service.py, routers/reports.py (REPORT_PAYLOAD_MAX_BYTES). Dual-track migration if the schema changes (SQLite db/migrations.py + Alembic). Coordinate with the export issue (streaming). OSS-core.