Field class should not let you analyze int values? [LUCENE-8108]

I stumbled on this by accident, by creating a `Field` instance with a `Integer` value for its `fieldsData` and then setting `tokenized = true` in its `FieldType`.

If you do this then Lucene silently converts the int to a string and then tokenizes it, e.g. applying synonyms, etc., if that's what your analysis chain does.

---
Migrated from [LUCENE-8108](https://issues.apache.org/jira/browse/LUCENE-8108) by Michael McCandless (@mikemccand), updated Dec 28 2017
Attachments: [LUCENE-8108.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-8108/LUCENE-8108.patch)

Start by reproducing the issue with a Field whose fieldsData is an Integer and a FieldType with tokenized=true. Trace the Field analysis path to determine where the integer becomes a string, then verify that integer values are no longer silently analyzed, including behavior involving synonyms.

Written by the indexing model from the issue text.