Pydantic-ify `RunConfig` and nested config dataclasses

[jc-macdonald]

Summary

Convert the frozen config dataclasses in core_solver.py to Pydantic BaseModel:

• RunConfig (L194)
• DtControllerConfig (L148)
• AdaptiveConfig (L162)
• OperatorSpecs (L177)

Motivation

These are user-facing configuration objects with no validation at construction time. Validation currently lives downstream in CoreSolver._resolve_run_plan(), meaning invalid configs only surface at solve time. Pydantic adds:

• Field constraints — method validated against MethodName literal (currently str), numeric bounds (ge=0, gt=0) on tolerances, safety factors, dt limits.
• Serialization — YAML/JSON round-tripping for config persistence and reproducibility.
• Immutability — model_config = ConfigDict(frozen=True) replaces @dataclass(frozen=True).
• Consistency — Matches the flepimop2 provider config (OpEngineEngineConfig) which is already Pydantic.

Scope

• Convert 4 dataclasses → Pydantic BaseModel with frozen=True
• Add field validators / constraints matching current downstream checks
• Update _resolve_run_plan() to rely on config-level validation
• Update all tests constructing these objects

Constraints

• Must not change RunPlan (internal-only, never user-constructed)
• Must not change mutable parameter bundles (StepIO, ImplicitStageParams, etc.) — hot-path structures
• Must pass ruff check --preview and mypy --strict

Related

• Complements Move validation logic from CoreSolver to RunConfig/RunPlan #31 (move validation from CoreSolver to RunConfig)
• Resolve CoreSolver mutable+return state pattern #32 (mutable-state cleanup)