feat: data versioning and result history

[tnunamak]

Context

Currently each source stores only the latest result at ~/.dataconnect/results/{source}.json. For diffing, trend analysis, and the vana next feature, we need historical snapshots.

Design considerations

• Unbounded disk growth: need a retention policy (e.g., keep last N results, or last N days)
• Storage format: timestamped files (github-2026-03-17.json) or a single file with version array
• Compression: results can be large (ChatGPT conversations). Consider gzip or delta encoding.
• Migration: existing last-result.json and per-source results should be preserved

Scope

This is infrastructure for vana context --since and vana next. Not needed for the initial skill-based approach where the agent parses timestamps from the current snapshot.