✨ Add language-agnostic DynamoDB schema definition (schema.yaml)

Problem or Use Case

The current DynamoDB schema is defined implicitly in Python code (schema.py and repository.py). This makes it difficult to:

1. Implement alternative aggregators (e.g., Rust) that need to read/write the same data
2. Generate type-safe clients in other languages
3. Validate schema consistency across implementations
4. Document the schema in a language-neutral format

As the project moves toward alternative backends (v1.4.0) and multi-language ecosystem growth, a single source of truth for the DynamoDB schema becomes essential.

Proposed Solution

Create a schema.yaml (or schema.json) file that defines:

1. Table structure: PK/SK patterns, GSI definitions
2. Record types: All item types with their attributes and types
3. Key builders: Documented patterns for constructing keys
4. Attribute constraints: Required fields, enums, value ranges

# Example structure
table:
  name: rate_limits
  billing_mode: PAY_PER_REQUEST
  
keys:
  primary:
    partition_key: PK
    sort_key: SK
  gsi1:
    partition_key: GSI1PK
    sort_key: GSI1SK
  gsi2:
    partition_key: GSI2PK
    sort_key: GSI2SK

prefixes:
  entity: "ENTITY#"
  parent: "PARENT#"
  resource: "RESOURCE#"
  system: "SYSTEM#"
  audit: "AUDIT#"

record_types:
  entity_meta:
    pk_pattern: "ENTITY#{entity_id}"
    sk: "#META"
    attributes:
      entity_id: { type: string, required: true }
      parent_id: { type: string, required: false }
      cascade: { type: boolean, default: false }
      created_at: { type: integer, description: "Unix timestamp ms" }
  
  composite_bucket:
    pk_pattern: "ENTITY#{entity_id}"
    sk_pattern: "#BUCKET#{resource}"
    attributes:
      # Composite bucket fields use b_{limit}_{field} naming
      "b_{limit}_tk": { type: integer, description: "Tokens (millitokens)" }
      "b_{limit}_cp": { type: integer, description: "Capacity (millitokens)" }
      # ...

Alternatives Considered

1. OpenAPI/JSON Schema: More verbose, less suited for DynamoDB-specific patterns
2. Protobuf: Adds compilation step, not as readable
3. Keep Python-only: Blocks multi-language growth

Acceptance Criteria

• schema.yaml file exists at project root (or docs/schema.yaml)
• Schema defines all record types currently in repository.py
• Schema includes all key prefixes and builders from schema.py
• Schema documents attribute types matching Python type hints
• Unit test validates Python implementation matches schema.yaml
• CLAUDE.md updated to reference schema.yaml as schema source of truth

Start by reading schema.py and repository.py to inventory the current DynamoDB keys, record types, prefixes, builders, and Python type hints. Add schema.yaml at the project root or under docs, then add a unit test comparing the Python implementation with it and update CLAUDE.md to identify the schema as the source of truth.