arrow

ARROW-4679: [Rust] Implement in-memory data source for DataFusion

This PR adds a new memory-based table implementation and updates the DataFusion benchmarks to load CSV data into memory before the benchmark starts and then measure the performance of executing queries against in-memory data.

This PR also renames the data source traits and organizes them better in a new namespace.

Finally, it is now possible to register arbitrary data sources with the execution context.

Author: Andy Grove <andygrove73@gmail.com>

Closes #3754 from andygrove/ARROW-4679 and squashes the following commits:

8b4c050 <Andy Grove> update benchmark
5ce48e4 <Andy Grove> improve tests
f1916a1 <Andy Grove> Add validation for projection in MemTable
5e8ddad <Andy Grove> Add schema validation when creating MemTable
2c72f08 <Andy Grove> Add check for batch column count, split test into two with better names
afe1c06 <Andy Grove> benchmarks now run against in-memory data
9b2136d <Andy Grove> update rustdoc
30c2520 <Andy Grove> rename traits
379d221 <Andy Grove> Implement in-memory data source for DataFusion