apacheGH-48112: [C++][Parquet] Use more accurate data length estimate when decoding PLAIN BYTE_ARRAY data

Author: pitrou

cpp/src/parquet/decoder.cc

@@ -790,8 +790,13 @@ class PlainByteArrayDecoder : public PlainDecoder<ByteArrayType> {
       return Status::OK();
     };
 
-    return DispatchArrowBinaryHelper<ByteArrayType>(out, num_values, len_,
-                                                    visit_binary_helper);
+    // We're going to decode up to `num_values - null_count` PLAIN values,
+    // and each value has a 4-byte length header that doesn't count for the
+    // Arrow binary data length.
+    int64_t estimated_data_length =
+        std::max<int64_t>(0, len_ - 4 * (num_values - null_count));
+    return DispatchArrowBinaryHelper<ByteArrayType>(
+        out, num_values, estimated_data_length, visit_binary_helper);
   }
 
   template <typename BuilderType>