ARROW-13344: [R] Initial bindings for ExecPlan/ExecNode

[nealrichardson]

This PR adds support for both scalar and group-by aggregation via dplyr::summarize(). Only the functions sum, any, and all are wired up. Followup issues (both bugs and features):

• [C++] Aggregation nodes seem not to respect FunctionOptions, or else I'm not passing them in correctly (ARROW-13497)
• [C++] ScanNode takes filter but doesn't filter (ARROW-13498)
• [R] Aggregation on expression doesn't NSE correctly (ARROW-13499)
• [R] Bindings for mean, var, sd aggregation (ARROW-13528)
• [R] Bindings for count aggregation (ARROW-13501)
• [R] Bindings for min/max aggregation (ARROW-13502)
• [R] Handle summarize() with 0 arguments or no aggregate functions (ARROW-13543)
• [R] Support .groups argument to summarize() (ARROW-13550)
• [C++] MakeScalarAggregateNode and MakeGroupByNode have quite different function signatures, which makes working with the API confusing; GroupBy doesn't let you specify the names of the output columns (ARROW-13482)
• [C++] Grouped aggregation functions all have to be invoked with a hash_ prefix to the name, which seems unnecessary because you can't call a non-hash-aggregation function in GroupBy and you can't call a hash_ function in ScalarAggregate (ARROW-13451)

[github-actions]

https://issues.apache.org/jira/browse/ARROW-13344

[bkietz]

• MakeScalarAggregateNode and MakeGroupByNode have quite different function signatures, which makes working with the API confusing
• GroupBy doesn't let you specify the names of the output columns

ARROW-13482 (#10793) unifies the factories for scalar and grouped aggregation. I'll ensure that PR also provides configurable output column names for both

• Grouped aggregation functions all have to be invoked with a hash_ prefix to the name, which seems unnecessary because you can't call a non-hash-aggregation function in GroupBy and you can't call a hash_ function in ScalarAggregate

ARROW-13451 would allow grouping of grouped and scalar aggregation kernels under the same function (rather than requiring distinct sum and hash_sum functions), which I think would address these