GH-2972: Fix incomplete avro metadata on INT96 schema converter

[ArnavBalyan]

Rationale for this change

• INT96 fields use a hardcoded "INT96" string as schema name.
• Avro re-uses the same schema reference on reoccurrence, which causes subsequent values as bare "INT96" instead of the complete definition.
• Move away from the hardcoding and use name/namespace map to prevent collisions.
• Fixes: AvroSchemaConverter toString() conversion when schema has multiple INT96 fields #2972.

Before:

{
  "type" : "record",
  "name" : "int96Schema",
  "fields" : [ {
    "name" : "timestamp_1",
    "type" : [ "null", {
      "type" : "fixed",
      "name" : "INT96",
      "doc" : "INT96 represented as byte[12]",
      "size" : 12
    } ],
    "default" : null
  }, {
    "name" : "timestamp_2",
    "type" : [ "null", "INT96" ],
    "default" : null
  } ]
}

After:

{
  "type" : "record",
  "name" : "int96Schema",
  "fields" : [ {
    "name" : "timestamp_1",
    "type" : [ "null", {
      "type" : "fixed",
      "name" : "timestamp_1",
      "doc" : "INT96 represented as byte[12]",
      "size" : 12
    } ],
    "default" : null
  }, {
    "name" : "timestamp_2",
    "type" : [ "null", {
      "type" : "fixed",
      "name" : "timestamp_2",
      "doc" : "INT96 represented as byte[12]",
      "size" : 12
    } ],
    "default" : null
  } ]
}

Are these changes tested?

• Yes

Are there any user-facing changes?

• Yes

Closes: #2972

[ArnavBalyan]

cc @gszadovszky : )

[ArnavBalyan]

cc @gszadovszky could you please merge it if all looks good thanks!

[wgtmac]

Merged it. Thanks @ArnavBalyan for working on this and @gszadovszky for the review!