[Format] Add AdbcStatementExecuteSchema

[lidavidm]

It would be useful to be able to get the schema of a query's result set without actually executing it. For example, Ibis uses this to allow embedding arbitrary SQL within its DSL. And Flight SQL already supports this.

We could add a new function:

AdbcStatusCode AdbcStatementExecuteSchema(
  struct AdbcStatement* statement,
  struct ArrowSchema* schema,
  struct AdbcError* error);

• Drivers may implement this by executing the query.
• They may attempt to transform the query to limit side effects and execution time, however, this is not guaranteed.
  • In other words, don’t use this with DML/DDL queries, though that doesn’t really make sense in the first place.
  • If the client itself is able to transform the query to limit execution time, this is still a good idea.

This applies across languages. See #317 for how backwards compatibility would be managed.

[tokoko]

I started rewriting spark-flight-sql data source to spark-adbc (https://github.com/tokoko/spark-adbc), this function would greatly simplify efficient schema inference. I can give implementing this a try (at least in Java) if that's okay with you.

P.S. Wouldn't it make more sense for the drivers to fail if schema detection can't be accomplished without significant execution time? (up to drivers to decide what "significant" means) I'm thinking of stored procedures in case of jdbc, for example. It would force clients to explicitly call executeQuery in such cases and avoid inadvertently executing long-running queries.

[lidavidm]

Failing probably makes more sense, we should make sure to define the error code in that case.

Since this affects the spec, we're still deciding how best to handle the development process; if you're interested, please chime in on the mailing list. https://lists.apache.org/thread/3y73755p0w4n25rd5ry2ppnhzkz3thv3