Add heuristic for detecting string dtype -> text in Elasticsearch instead of always keyword

When calling `pandas_to_eland()` it'd be good to start creating `text` fields if data is:

- Longer than 255 characters (longer than default keyword)
- Contains 5+ spaces (ML magic number)
- Has a high cardinality of values compared to entries (This one is hard to detect)

Start by locating pandas_to_eland() and inspect its current string-to-Elasticsearch mapping behavior. Define and test how string length, whitespace, and cardinality should influence text versus keyword selection, including the unspecified high-cardinality case. Done means the behavior is covered and pandas data produces the intended Elasticsearch mapping.

Written by the indexing model from the issue text.