[C++][Parquet] CleanStatistic for ByteArray might prune the zero-sized array

[mapleFU]

Describe the bug, including details regarding any error messages, version, and platform.

The code[1] would cleanup the min-max stats in Parquet. For ByteArray, we may "Merge" multiple stats when reading from file. Things would be tricky in the code below when min = ""

1. Code in [2] is empty, so PlainDecode will not be called, and has_min_max_ is true. But ByteArray keeps default constructor, which leaves ptr == nullptr [3]
2. When call TypedStatistics::Merge, this will call Cleanup [1], and finally, the min-max statistics would leave unchanged.

So, when min = "" being merged, the min-max will keep the old statistics.

[1]

arrow/cpp/src/parquet/statistics.cc

Line 408 in ea47172

optional<std::pair<ByteArray, ByteArray>> CleanStatistic(

[2]

arrow/cpp/src/parquet/statistics.cc

Line 609 in ea47172

if (!encoded_min.empty()) {

[3]

arrow/cpp/src/parquet/types.h

Line 587 in ea47172

ByteArray() : len(0), ptr(NULLPTR) {}

Component(s)

C++, Parquet