GH-47596: [C++][Parquet] Fix printing of large Decimal statistics (#47619)

### Rationale for this change

Parquet CLI tools fail printing the statistics for a Decimal column with a precision larger than the max Decimal128 precision.

Example:
```console
$ /build/build-test/debug/parquet-reader --only-metadata /tmp/pqfuzz/pq-table-1
...
Column 5: col_6 (FIXED_LEN_BYTE_ARRAY(11) / Decimal(precision=24, scale=7) / DECIMAL(24,7))
Column 6: col_7 (FIXED_LEN_BYTE_ARRAY(18) / Decimal(precision=43, scale=7) / DECIMAL(43,7))
...
Column 5
  Values: 375, Null Values: 74, Distinct Values: 0
  Max (exact: true): 98505381700645007.0205463, Min (exact: true): -99708959786297168.1726196
  Compression: UNCOMPRESSED, Encodings: PLAIN(DICT_PAGE) RLE_DICTIONARY
  Uncompressed Size: 3754, Compressed Size: 3754
Column 6
  Values: 375, Null Values: 69, Distinct Values: 0
  Max (exact: true): Parquet error: Failed to parse decimal value: Length of byte array passed to Decimal128::FromBigEndian was 18, but must be between 1 and 16
...
```

### What changes are included in this PR?

Use Decimal256 instead of Decimal128 when printing a Decimal statistic.

### Are these changes tested?

Yes, by new tests.

### Are there any user-facing changes?

No.
* GitHub Issue: #47596

Authored-by: Antoine Pitrou <[email protected]>
Signed-off-by: Antoine Pitrou <[email protected]>

Author: pitrou

cpp/src/parquet/types.cc

@@ -137,7 +137,7 @@ std::string FormatDecimalValue(Type::type parquet_type, ::std::string_view val,
     }
     case Type::FIXED_LEN_BYTE_ARRAY:
     case Type::BYTE_ARRAY: {
-      auto decimal_result = ::arrow::Decimal128::FromBigEndian(
+      auto decimal_result = ::arrow::Decimal256::FromBigEndian(
           reinterpret_cast<const uint8_t*>(val.data()), static_cast<int32_t>(val.size()));
       if (!decimal_result.ok()) {
         throw ParquetException("Failed to parse decimal value: ",

cpp/src/parquet/types_test.cc

@@ -138,6 +138,8 @@ TEST(TypePrinter, StatisticsTypes) {
   ASSERT_EQ("0x696a6b6c6d6e6f70", FormatStatValue(Type::FIXED_LEN_BYTE_ARRAY, smax));
 
   // Decimal
+
+  // If the physical type is INT32 or INT64, the decimal storage is little-endian.
   int32_t int32_decimal = 1024;
   smin = std::string(reinterpret_cast<char*>(&int32_decimal), sizeof(int32_t));
   ASSERT_EQ("10.24", FormatStatValue(Type::INT32, smin, LogicalType::Decimal(6, 2)));
@@ -147,7 +149,8 @@ TEST(TypePrinter, StatisticsTypes) {
   ASSERT_EQ("10240000.0000",
             FormatStatValue(Type::INT64, smin, LogicalType::Decimal(18, 4)));
 
-  std::vector<char> bytes = {0x11, 0x22, 0x33, 0x44};
+  // If the physical type is BYTE_ARRAY or FLBA, the decimal storage is big-endian.
+  std::vector<uint8_t> bytes = {0x11, 0x22, 0x33, 0x44};
   smin = std::string(bytes.begin(), bytes.end());
   ASSERT_EQ("28745.4020",
             FormatStatValue(Type::BYTE_ARRAY, smin, LogicalType::Decimal(10, 4)));
@@ -156,6 +159,19 @@ TEST(TypePrinter, StatisticsTypes) {
   ASSERT_EQ("0x11223344", FormatStatValue(Type::BYTE_ARRAY, smin));
   ASSERT_EQ("0x11223344", FormatStatValue(Type::FIXED_LEN_BYTE_ARRAY, smin));
 
+  bytes = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+           0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff,
+           0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xcf, 0xc7};
+  smin = std::string(bytes.begin(), bytes.end());
+  ASSERT_EQ("-12.345",
+            FormatStatValue(Type::BYTE_ARRAY, smin, LogicalType::Decimal(40, 3)));
+  ASSERT_EQ("-12.345", FormatStatValue(Type::FIXED_LEN_BYTE_ARRAY, smin,
+                                       LogicalType::Decimal(40, 3)));
+  ASSERT_EQ("0xffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffcfc7",
+            FormatStatValue(Type::BYTE_ARRAY, smin));
+  ASSERT_EQ("0xffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffcfc7",
+            FormatStatValue(Type::FIXED_LEN_BYTE_ARRAY, smin));
+
   // Float16
   bytes = {0x1c, 0x50};
   smin = std::string(bytes.begin(), bytes.end());