Add better and faster support for dictionary types

[alamb]

Note: migrated from original JIRA: https://issues.apache.org/jira/browse/ARROW-8464

Usecases: Efficiently process large columns of low cardinality Strings
 

• BatchIterator should accept both DictionaryBatch and RecordBatch
• Type Coercion optimizer rule should inject expression for converting dictionary value types to index types (for equality expressions, and IN(values, ...)
• Use specialized dictionary compute kernels for binary PhysicalExpr evaluation #1178

[alamb]

Comment from Andrew Lamb(alamb) @ 2020-10-06T12:25:16.639+0000:

FYI [~andygrove] -- I am doing some part of this in ARROW-10159 -- however, the initial implementation effectively converts DictionaryArray --> PrimitiveArray / StringArray and then uses the existing processing.

To support the actual efficient processing usecase, I think significant work will be needed to add appropriate dictionary support to the arrow compute kernels