Replace options dict with a dataclass

[jc-macdonald]

Summary

Replace the 30-key raw dict used by _build_options() / _options() with a @dataclass.

Current pattern

opts_default = {
    "init": "random",
    "maxiters": 1000,
    "bias": True,
    "hp_va": 0.001,
    # ... 25+ more string keys
}
opts, wrnmsg = _options(opts_default, **kwargs)

Case-insensitive key matching in _options.py works around unclear contracts. Callers pass and receive untyped dicts with no IDE support.

Proposed change

@dataclass
class PCAOptions:
    init: str = "random"
    maxiters: int = 1000
    bias: bool = True
    hp_va: float = 0.001
    hp_vb: float = 0.001
    hp_v: float = 0.001
    # ... rest as typed fields

    @classmethod
    def from_kwargs(cls, **kwargs: object) -> PCAOptions:
        return cls(**{k: v for k, v in kwargs.items() if hasattr(cls, k)})

Follows the pattern already used by SelectionConfig in model_selection.py.

Benefits

• Construction-time validation (typos fail immediately)
• IDE autocomplete for every option
• mypy catches type mismatches
• Eliminates the case-insensitive key hack

Scope

~80 lines of dataclass + update _options.py logic. Non-breaking — from_kwargs() classmethod preserves the existing **kwargs API.

Depends on

• Use str Enums for categorical options (algorithm, init, compat_mode) #91 (Enum types should be field types in the dataclass)