Add schema fingerprinting to Tool and Flow

[dgenio]

Context

Split from #22 (schema fingerprinting, drift detection, and flow status management). This issue focuses on the schema fingerprinting utility and Tool.schema_hash property.

What to do

1. Schema fingerprinting utility

Add a utility function to compute a deterministic hash of a Pydantic model's JSON Schema:

# chainweaver/compat.py
import hashlib, json

def schema_fingerprint(model: Type[BaseModel]) -> str:
    """Compute a SHA-256 fingerprint of a Pydantic model's JSON Schema."""
    schema = model.model_json_schema()
    canonical = json.dumps(schema, sort_keys=True, separators=(",", ":"))
    return hashlib.sha256(canonical.encode()).hexdigest()[:16]

2. Tool schema hash properties

Add computed properties to Tool:

class Tool:
    @property
    def input_schema_hash(self) -> str:
        return schema_fingerprint(self.input_schema)

    @property
    def output_schema_hash(self) -> str:
        return schema_fingerprint(self.output_schema)

    @property
    def schema_hash(self) -> str:
        """Combined hash of input + output schemas."""
        combined = self.input_schema_hash + self.output_schema_hash
        return hashlib.sha256(combined.encode()).hexdigest()[:16]

3. Flow-level schema snapshot

When a flow is compiled/registered with validation, store the tool schema hashes:

class Flow(BaseModel):
    # ... existing fields ...
    tool_schema_hashes: Optional[dict[str, str]] = None  # tool_name → schema_hash

4. Compatibility validator

def check_flow_compatibility(
    flow: Flow,
    tools: dict[str, Tool],
    expected_hashes: Optional[dict[str, str]] = None,
) -> list[CompatibilityIssue]:
    """
    Check that each step's tool exists and its schema matches expectations.
    Returns a list of issues (empty = compatible).
    """

Files to create/modify

• chainweaver/compat.py — new module with fingerprinting + validator
• chainweaver/tools.py — add hash properties to Tool
• chainweaver/flow.py — add tool_schema_hashes to Flow
• tests/test_compat.py — new test file

Acceptance Criteria

• schema_fingerprint(model) produces a deterministic, stable hash string
• Same model always produces the same fingerprint
• Different models (different fields, types, or constraints) produce different fingerprints
• Tool.schema_hash property exists and is computed from input + output schemas
• check_flow_compatibility() detects: missing tools, schema hash mismatches
• Flow.tool_schema_hashes stores a snapshot at registration/compile time
• At least 6 test cases: same schema → same hash, field added → different hash, field renamed → different hash, compatibility pass, compatibility fail, missing tool

Out of Scope

• Drift detection on tool re-registration (see split issue)
• Flow status management (see split issue)
• Automatic schema migration
• Semantic compatibility (e.g., int compatible with float)

Notes

• The 16-char hex prefix is sufficient for collision avoidance in practical scenarios.
• Pydantic v2's model_json_schema() output is stable for the same model definition — but beware of Python version differences in dict ordering (mitigated by sort_keys=True).
• This pairs with Add flow and tool schema versioning with compatibility checks #15 (versioning) — schema hashes are the mechanism behind version compatibility checks.

---

Split from #22. See also: Flow status management issue, Schema drift detection issue.