Logic around DEFAULT_ANALYZER is a bit off

[joestump]

Took quite a while for me to figure out why I was getting "analyzer": None for some of my mappings. This line for setting the analyzer and this line from ConfigurableFieldMixin conspired to make this happen. That getattr will always return true because the kwargs.pop defaults to None and sets the attribute. The default setting is never triggered when this happens, which results in field_class.analyzer being a valid attribute set to None.

I think one of two things should probably happen:

• Change the logic so that it checks if getattr(field_class, 'analyzer', self.DEFAULT_ANALYZER) is None and, if so, set it to self.DEFAULT_ANALYZER.
• Raise a ValueError or haystack.exceptions.SearchFieldError if someone doesn't provide analyzer in the CharField's kwargs.

If you let me know which one you think is correct, I'll try and get a PR wrapped up soon.

[joestump]

TIL: If you set analyzer to null, Elasticsearch will remove all analyzer and index settings from the field mappings configuration on the server.

[bennylope]

In this case I think the most sensible approach would be to simply raise an exception is not analyzer is provided.