Add Arrow dictionary and other optimizations.

[jheer]

• Add an internal dictionary column abstraction.
• Add "pay-as-you-go" extractors for nested Arrow types, rather than unpacking.
• Add inlined table scans in performance critical sections (includes derive, filter, groupby, and aggregate).
• Update expression parser to optimize equality checks for dictionary/literal comparisons.
• Update expression parser to annotate direct column and operator references.
• Update groupby() verb to directly use dictionary keys when applicable.
• Update rollup() logic to generate table-compatible column arrays and assign them when applicable.
• Add performance benchmarks for direct Arrow and Arrow-Arquero operations.
• Fix AST tree walker bug for list types.

Close #86.