feat: Python SDK stub for one-line ML framework integration

[joelpeace48-cell]

Summary

Adds the ModelTraceClient Python SDK stub, enabling ML engineers to integrate ModelTrace into existing training and inference pipelines with minimal code changes.

API Design

from modeltrace import ModelTraceClient

client = ModelTraceClient(
    rpc_url="https://soroban-testnet.stellar.org",
    secret_key="S...",
    contract_id="C..."
)

model_id = client.register_model(
    name="MyLLM-v1",
    version="1.0.0",
    weights_hash="sha256:abc123...",
    training_data_cid="QmXzY7bN9..."
)

event_id = client.log_inference(
    model_id=model_id,
    payload_hash="sha256:def456...",
    risk_level="LOW"
)

What's included

• ModelTraceClient class with fully typed method signatures
• ModelRecord, AuditEvent, RegistryStats dataclass stubs
• RiskLevel literal type for type-safe risk classification
• Documented integration points for stellar-sdk Soroban transaction builder
• Explicit NotImplementedError with helpful guidance (not silent failures)

What's next

• Implement Soroban transaction builder in register_model()
• Implement log_inference() with async support
• PyPI package modeltrace-sdk
• MLflow autolog integration
• Hugging Face Hub hook

Test plan

• Unit tests pending stellar-sdk integration
• Integration test against testnet deployment

Related: #14 (Python SDK milestone)