[WIP] [format] Support narrowing Parquet reads -- do not merge this!

[leaves12138]

Purpose

Some externally produced Parquet files can store values with a wider physical type than the Paimon table schema expects. For example, an INT field may be stored as Parquet INT64 and a FLOAT field may be stored as Parquet DOUBLE. When such columns are dictionary encoded, the vectorized reader currently selects the updater only from the Paimon logical type and calls methods such as decodeToInt, which fails for dictionaries such as PlainLongDictionary.

Changes

• Use a narrowing updater when reading Paimon INT from Parquet INT64.
• Use a narrowing updater when reading Paimon FLOAT from Parquet DOUBLE.
• Keep overflow checks for INT64 -> INT via Math.toIntExact.
• Add regression tests for INT64 -> INT and DOUBLE -> FLOAT reads.

Tests

• mvn -pl paimon-format -am -Pfast-build -DfailIfNoTests=false -DwildcardSuites=none -Dtest=FileTypeNotMatchReadTypeTest test