Canonical ClinicalPrediction model + Delta outbox schema

[slusset]

Parent Epic

Part of #16 — [Epic] Databricks EHR Writeback — Delta Outbox + Canonical Prediction Model

Summary

Define the canonical ClinicalPrediction domain model and the Delta outbox table that stores it with processing semantics.

Tasks

• Define ClinicalPrediction Pydantic model with fields: prediction_id, patient_id, encounter_id, prediction_code, values[] (label, value, unit), effective_datetime, model_id, model_version, metadata
• Define PredictionValue Pydantic model for the values array (supports scalar and vector)
• Define Delta outbox table schema — adds processing columns: status (pending/processing/completed/failed/dead_letter), locked_by, locked_at, retry_count, last_error, created_at, updated_at
• Implement lease-based MERGE semantics: claim (pending → processing with lock), release (processing → completed/failed), expire stale locks
• New module: src/ehr_writeback/outbox/ with models.py, outbox_store.py
• Unit tests for model validation and outbox CRUD operations

Acceptance Criteria

• ClinicalPrediction model validates scalar (single value) and vector (multi-value) predictions
• Outbox store can claim a batch of N pending records atomically via MERGE
• Stale locks (older than configurable timeout) are automatically reclaimable
• Unit tests pass with mocked Spark session