[python] Add batch numeric extraction for Variant fields

[XiaoHongbo-Hope]

Purpose

Wide numeric Variant workloads currently materialize many Python dictionaries and scalar objects before building numeric arrays.

Changes

• Extract ordered, literal top-level numeric fields into Arrow FixedSizeList<Float32>.
• Reuse the Variant metadata and exact object layout within a batch while validating object offsets and selected child boundaries.
• Preserve SQL null rows and represent missing, Variant-null, or non-object-root fields as child nulls.
• Expose pypaimon_rust.data.variant_get_numeric_fields and release the GIL during extraction.
• Return NotImplementedError for non-numeric selected values so an integration can explicitly fall back.

Scope and trade-offs

• This reduces decoding and Python object materialization; it does not reduce Parquet or object-store I/O.
• Supported values are Variant integer, float, double, and decimal. Conversion to float32 may lose precision.
• The result follows the requested field order. Field names are literal, not Variant path expressions.
• Automatic PyPaimon selection and application SDK integration are intentionally outside this PR.

Tests

• cargo test -p paimon --lib variant (46 passed)
• pytest -q bindings/python/tests/test_variant.py (1 passed)
• cargo clippy -p paimon --lib -p pypaimon_rust -- -D warnings
• cargo fmt --all --check