chore: migrate analytics/calibration.py to Pydantic BaseModel — pilot (AC-489, AC-481)

[jayscambler]

Summary

Pilot migration of analytics dataclasses to Pydantic BaseModel, establishing the pattern for the remaining ~60 dataclasses with manual to_dict/from_dict.

Problem

295 dataclasses across the codebase implement manual to_dict() -> dict[str, Any] and from_dict(data) classmethods — boilerplate that Pydantic provides for free via model_dump() and model_validate().

Changes

Converted analytics/calibration.py (3 classes)

Class	to_dict lines removed	from_dict lines removed
CalibrationSample	15 → 1	15 → 1
CalibrationOutcome	11 → 1	11 → 1
CalibrationRound	8 → 1	10 → 1

Total: ~78 lines of boilerplate → 6 lines of aliases

Backward compatibility

to_dict() and from_dict() kept as thin aliases (self.model_dump() / cls.model_validate(data)) so existing callers (CalibrationStore, SpotCheckSampler) work unchanged.

Pattern for future migrations

# Before (dataclass)
@dataclass(slots=True)
class Foo:
    x: str
    y: int = 0
    
    def to_dict(self) -> dict[str, Any]:
        return {"x": self.x, "y": self.y}
    
    @classmethod
    def from_dict(cls, data: dict[str, Any]) -> Foo:
        return cls(x=data["x"], y=data.get("y", 0))

# After (Pydantic)
class Foo(BaseModel):
    x: str
    y: int = 0
    
    def to_dict(self) -> dict[str, Any]:
        return self.model_dump()
    
    @classmethod
    def from_dict(cls, data: dict[str, Any]) -> Foo:
        return cls.model_validate(data)

TDD

Added test_pydantic_migration.py with 5 tests:

• model_dump roundtrip
• JSON serialization
• Nested model handling
• Backward-compat to_dict alias

Verification

• ruff check src — all checks passed
• mypy src — zero errors
• pytest tests/ — all tests pass

Issues

Partially resolves AC-489 and AC-481

[linear]

AC-489 Eliminate 295 hand-rolled to_dict/from_dict methods — use Pydantic or dataclasses.asdict

AC-481 Standardize dataclass vs Pydantic usage — 328 dataclasses + 85 Pydantic models with no clear boundary