feat: add _numeric_sentinel_eligible dtype guard to _null_detection.py

[DEVunderdog]

Parent

#94

What to build

Add _numeric_sentinel_eligible(dtype) to utils/_null_detection.py — a pure dtype predicate that returns True for all integer and float numeric Polars dtypes and False for everything else.

This becomes the single authority for which columns are eligible to receive user-declared numeric sentinel normalization, mirroring the existing _sentinel_eligible (string) and _inf_eligible (float NaN/Inf) functions already in the same module.

Acceptance criteria

• _numeric_sentinel_eligible is defined in utils/_null_detection.py alongside the existing _sentinel_eligible and _inf_eligible functions
• Returns True for all Polars integer dtypes (Int8, Int16, Int32, Int64, UInt8, UInt16, UInt32, UInt64)
• Returns True for Float32 and Float64
• Returns False for String/Utf8, Boolean, Date, Datetime, Duration, and any other non-numeric dtype
• Unit tests added to tests/unit/profiling/test_null_detection.py covering all dtype cases above
• Docstring follows the numpy-style convention required by ADR-0034

Blocked by

None — can start immediately