Implement BinaryRow deserialization to decode partition binary bytes

[luoyuxia]

Parent Issue

Part of #124 (support partitioned table)

Background

In Paimon, ManifestEntry._PARTITION stores partition values as raw BinaryRow binary bytes (Vec<u8>). Currently, BinaryRow (crates/paimon/src/spec/data_file.rs:30-51) is a stub that only stores arity — it cannot hold or parse actual partition data.

To support partitioned table reading, we need BinaryRow to wrap the raw bytes and provide typed accessor methods.

What needs to be done

1. Enhance BinaryRow to hold actual binary data

   • Add a data: Vec<u8> (or Bytes) field to back the row
   • Parse the binary layout: header (8 bytes) + null bit set (aligned) + fixed-length part (8 bytes per field) + variable-length part

2. Implement typed getters

   • is_null_at(pos) — check the null bit
   • get_int(pos) → i32
   • get_long(pos) → i64
   • get_short(pos) → i16
   • get_byte(pos) → i8
   • get_boolean(pos) → bool
   • get_float(pos) → f32
   • get_double(pos) → f64
   • get_string(pos) → &str / String (variable-length, read offset+length from fixed part, then read from variable part)
   • get_binary(pos) → &[u8]

3. Add a constructor from raw bytes

   • BinaryRow::from_bytes(arity: i32, data: Vec<u8>) -> BinaryRow

4. Unit tests

   • Round-trip: construct binary data manually, verify typed getters
   • Test null handling
   • Test variable-length fields (String, Binary)

Reference

• Java: BinaryRow
• Java: BinaryRowSerializer

Affected files

• crates/paimon/src/spec/data_file.rs — BinaryRow struct