Followup to ARROW-10322. In ARROW-9187, where we started making use of more compute functions in R, we found a couple of places where implicit casts weren't being inserted where they should:

• 843ff2a#diff-79100695986bbd6a63704fe9f238ce3ae9a39ddd093b7f6b213d4a722309d20aR576 "Function multiply_checked has no kernel matching input types (scalar[double], array[int32])"

• 843ff2a#diff-79100695986bbd6a63704fe9f238ce3ae9a39ddd093b7f6b213d4a722309d20aR590 "Function add_checked has no kernel matching input types (array[double], array[int32])" because implicit casts are only applied to scalars to cast them to the type of the other argument

  This may speak to a need for more rules around how inputs should be casted/promoted in different contexts.

Reporter: Neal Richardson / @nealrichardson
Assignee: Ben Kietzman / @bkietz

Related issues:

• [C++] Add "DispatchBest" APIs to compute::Function that selects a kernel that may require implicit casts to invoke (duplicates)

_{Note: This issue was originally created as ARROW-11080. Please see the migration documentation for further details.}