feat: Define metric schema and TypeScript types

[charliewwdev]

Create shared types for MetricEvent, MetricSeries, and AggregatedMetric. Types must be shared between the server WebSocket emitter and the client consumer to prevent payload drift.

---

Generated by OpenDev AI

All types are derived from Zod schemas using z.infer<>, consistent with the existing prd-decompose.ts / task-codegen.ts pattern. This gives both compile-time safety and runtime validation on WebSocket frames — preventing payload drift because the same schema object is the single source of truth for both sides. Key decisions: (1) MetricEvent is the atomic unit; MetricSeries batches historical points on subscribe; AggregatedMetric carries pre-computed stats so clients stay thin. (2) All three are wrapped in a ServerMetricMessage discriminated union so client switch statements are exhaustive. (3) ClientMetricMessage mirrors the pattern for subscription control messages. (4) Refinements enforce invariants (window_end >= window_start, min <= max) at the schema level so invalid shapes are rejected at the boundary. (5) Three parse helpers are exported: strict parse* for startup-time validation, and safeParse* for high-throughput loops where a bad frame should be skipped rather than crashing.