feat(analytics): consolidate agent run, token, and cost observability

[esafwan]

Summary

Consolidates the complete execution analytics and usage-observability work from #445 into one reviewable PR.

Included from #445

• Persist compatibility-safe usage snapshots and cost provenance on terminal Agent Run writes.
• Add immutable hourly and daily Agent Run Analytics Rollup records.
• Refresh rollups every five minutes with a correction window.
• Add portable raw-run and rollup indexes through a post-model-sync migration.
• Add System Manager aggregate analytics APIs.
• Add an Executions-page KPI dashboard backed by scheduled rollups.
• Add prompt-cache tracking and cache economics indicators.
• Add context composition metrics and the Agent Run context bar.
• Preserve the original analytics screenshots and test modules.

Design boundaries

• The frontend reads precomputed rollup buckets and does not aggregate raw runs in the browser.
• Existing cached_tokens and cost compatibility fields remain intact.
• The authoritative usage snapshot is null-aware and retains cost provenance.
• Global rollups are initially System Manager scoped because ownership filtering is not yet safe.

Validation

• Python compilation passed for all changed backend modules.
• git diff --check passed.
• The source PR reported yarn typecheck, production yarn build, migration, analytics tests, and live hourly/daily rollup/API verification.
• Frontend checks should be rerun by CI for this consolidation branch.

Source PR

• Supersedes feat: add scheduled agent run analytics #445.

Follow-up

• Preserve the detailed cache/cost analytics rationale and live-probe notes in the CacheAnalytics track documentation.