feat: extend _resolve_effective_nulls to normalize user-declared numeric sentinels

[DEVunderdog]

Parent

#94

What to build

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

For each column present in both the DataFrame and the sentinels dict, if the column's dtype passes _numeric_sentinel_eligible, append Polars expressions that replace each matching sentinel value with pl.lit(None, dtype=dtype). The function remains pure and side-effect-free; sentinel expressions are appended to the existing expression list alongside the existing string-sentinel and Inf/NaN rules.

When numeric_sentinels is None or {}, the function is strictly identical to its current behaviour — no existing caller is broken.

Acceptance criteria

• _resolve_effective_nulls signature is (df, numeric_sentinels=None)
• Int64 column: declared sentinel value converted to null; non-sentinel values unchanged
• Float64 column: declared sentinel value converted to null; existing NaN/Inf rules still apply independently in the same call
• Multiple sentinels declared for one column: all converted; any other value unchanged
• Sentinel declared for a column not present in the DataFrame: no error, rest of DataFrame unchanged
• Sentinel value that matches no rows: DataFrame unchanged (no-op, no error)
• Column with no declared sentinel: completely unchanged even if its values happen to equal a sentinel declared for another column
• numeric_sentinels=None: output is 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: add _numeric_sentinel_eligible dtype guard to _null_detection.py #170