feat: extend _resolve_effective_nulls with user-declared string sentinel replace semantics

[DEVunderdog]

Parent

#176

What to build

Extend _resolve_effective_nulls in utils/_null_normalization.py to accept an optional string_sentinels: dict[str, list[str]] | None parameter (default None).

For each String/Utf8 column in the DataFrame:

• If the column name is present in string_sentinels, match only the declared values (case-insensitive via .str.to_uppercase()). The hardcoded _SENTINEL_STRINGS set is not applied for that column.
• If the column name is absent, fall back to the existing _SENTINEL_STRINGS behaviour (no change).
• Empty/whitespace-only string detection is always applied unconditionally, regardless of any declaration.

The function remains pure and side-effect-free. When string_sentinels is None or {}, behaviour is strictly identical to the current implementation — no existing caller is broken.

Acceptance criteria

• _resolve_effective_nulls signature gains string_sentinels: dict[str, list[str]] | None = None
• String/Utf8 column with no declaration → hardcoded _SENTINEL_STRINGS applies (regression: current behaviour preserved)
• String/Utf8 column with user-declared sentinels → only declared values converted to null; a hardcoded default not in the declaration (e.g. "?") is not converted
• Declared sentinel matching is case-insensitive
• Multiple declared sentinels for one column → all converted; any other value unchanged
• Empty/whitespace strings always converted to null regardless of whether a declaration exists for that column
• Column present in string_sentinels but absent from the DataFrame → no error, rest of DataFrame unchanged
• string_sentinels=None → output byte-for-byte identical to calling with no argument
• All cases above covered by unit tests in tests/unit/imputation/test_null_normalization.py
• Docstring updated per ADR-0034

Blocked by

• Blocked by feat: ProfileConfig.string_sentinels declaration + StructuralProfileResult propagation #177