Add per-dog distraction analytics and event annotations

[OwenTanzer]

Summary

• add per-dog distraction summaries using observed ordinal distributions and median severity categories
• add a mobile-friendly timeline that displays one selected distraction category at a time
• preserve missing-data semantics by charting only explicitly logged observations, including explicitly recorded Absent values
• add persistent dated dog events with create, edit, and delete controls
• render dog events as vertical contextual annotations on the distraction timeline
• remove a dog's contextual events when the dog is deleted and backfill the new collection for existing databases

Why

Issue #55 asks for longitudinal distraction analytics without inventing zero-value observations or introducing false precision. A categorical distribution plus an observed median keeps the scale ordinal, while the single-series selector keeps the chart readable on small screens. Dated event annotations give trainers context for changes in behavior over time.

Data and compatibility

• adds dogEvents to the shared database blob and Worker default blob
• normalizes older persisted blobs that do not contain dogEvents
• stores domain dates separately from audit timestamps
• requires non-empty labels and valid calendar dates

Validation

• npm test — 28 tests passed, including ordinal aggregation and no-fabricated-observation coverage
• npm run build — production TypeScript and Vite build passed
• npm run lint — passed with 3 pre-existing warnings
• npm run typecheck in worker/ — passed
• git diff --check — passed

Closes #55