I concur with Dan’s year-old TODO: DistLoader doesn’t seem like
a useful abstraction and we’re better off just getting an unwrapped
DistSQLPlanner into the caller’s hands so they can use its public API
without fighting though additional indirection.

Indeed, Ideally I’d like to follow this with exporting more of the
DistSQL API, with the goal of eventually moving LoadCSV method to a
function in importccl defined entirely in terms of the public DistSQL
API, though that will obviously require some refactoring and careful
thinking about where to draw maintainable lines between these packages.

Release note: none.

This adds a DistSQL processor that writes whatever it is given as CSV
files to a specified storage location.

This processor can be appended as a sink for arbitrary SELECT queries.

The primary intended use-case is dumping tables as CSV for bulk-export,
but it can also be used on any SELECT query, so it could also be used in
report generation or other workflows.

Release note (enterprise change): prototype support for EXPORT CSV.