Fix namespace generation in Avro schemas

[acmiyaguchi]

This PR, along with mps#291 and mps#290 fixes generation of Avro schemas.

• Fixed the type of record fields. They were originally flattened to look like this:

  {
  "type":"record",
  "fields":[
    {
       "name":"foo",
       "type":"record",
       "fields":[
          {
             "name":"bar",
             "type":"null"
          }
       ]
    }
  ]
  }

  They should actually be in the following simplified form:

  {
  "type":"record",
  "fields":[
    {
       "name":"foo",
       "type": {
           "name": "foo",
           "type":"record",
           "fields":[
              { 
                 "name":"bar",
                 "type":"null"
              }
           ]
         }
    }
  ]
  }

  However, this forces the more verbose form of simple types due to serialization.

• Fixed union types by using vectors directly e.g. {"type": ["null", "int"]} to ["null", "int"].

• Support de-serialization of tuple-types in jsonschema, but not conversion.

• Add support for inferring namespaces in Tag::infer_name

After running the ./test-mps-valid-avro.py script, all of the generated schemas should now be valid avro.