[spark] Convert EqualNullSafe(NaN) to isNaN in V2 filters

[cxzl25]

Purpose

Tests

Added V2Filter: EqualNullSafe NaN to SparkV2FilterConverterTestBase.

[JingsongLi]

This has end-to-end value: Spark's f <=> NaN must retain NaN rows through Paimon pruning. Using IsNaN matches the existing = conversion; its row evaluation rejects NULL and its statistics evaluation remains conservative. I also checked the NOT path: IsNaN.negate() is unsupported, so conversion falls back instead of introducing incorrect negated pushdown.

No blocking issue found in this focused source review. The added FLOAT/DOUBLE SQL cases cover the intended result and nonempty scan. I have not rerun the Spark suite locally.