AVRO-3966: [Java] Fix default value serialisation for fixed and bytes

[mitasov-ra]

What is the purpose of the change

Default value for fixed and bytes schemas is converted to JSON in the wrong (according to docs) way.

Also, it's possible to store any unicode string as default for bytes, like:

{
  "type" : "record",
  "name" : "TestRecord",
  "fields" : [ {
    "name" : "testFixed",
    "type" : {
      "type" : "fixed",
      "name" : "Code",
      "size" : 3
    },
    "default" : "Привет"
  } ]
}

The above example would be serialised into Avro format as "?????".

In this PR:

• added explicit conversion from String to byte[] using ISO_8859_1 charset (the one that used during byte[] to JSON conversion and during reading Avro into Java objects);
• explicitly stated in doc-blocks of methods bytesDefault and fixedDefault that this conversion is performed;
• byte[] to JSON conversion itself now is done according to doc: by translating bytes to u-escaped sequences (e.g. [2F, 00, 11, FF] into \u002F\u0000\u0011\u00FF)

Verifying this change

This change is already covered by existing tests, such as:

• org.apache.avro.generic.TestGenericData
• org.apache.avro.TestFixed

DISCUSSION IS NEEDED

In the test TestGenericData I've had to rewrite test toStringEscapesControlCharsInBytes. The new behaviour encodes "a\nb" as "\u0061\u000A\u0062", thus breaking some compatibility (kinda, more like breaking readability).

It is possible still to not encode ASCII characters (from U+0020 to U+007E) and persist full compatibility with previous version, but then the behaviour would differ from one described in the documentation.

Documentation

• Does this pull request introduce a new feature? No, it just makes old feature work exactly how it's documented.

[mitasov-ra]

@Fokko there's an issue with checks you've triggered, I can't see logs of failed runs. Can you please rerun one more time?