GH-797: [JDBC] Fix PreparedStatement#execute for DML/DDL

[ennuite]

What's Changed

Instead of always obtaining a result set for queries issued via
PreparedStatement.execute(), we now inspect the dataset_schema
returned in ActionCreatePreparedStatementResult. If the schema has no
fields, we retrieve the update count instead. This aligns the return
value with the expectations of the JDBC API.

For such cases, the Arrow Flight SQL path now uses
CommandPreparedStatementUpdate instead of
CommandPreparedStatementQuery. This change mirrors the existing
approach in Statement.execute() and Statement.executeUpdate().

Are these changes tested?

Yes

Closes #797.

[lidavidm]

Thanks, this looks reasonable to me.

[lidavidm]

@alamb @jbonofre do InfluxDB and Dremio satisfy the assumptions here (dataset_schema will be populated if and only if the query of a prepared statement returns a result set)? I'm wondering if there might be systems that don't analyze the query at first and always omit the schema until actual execution

(We should really try to set up integration testing with Dremio and InfluxDB...)

[alamb]

@alamb @jbonofre do InfluxDB and Dremio satisfy the assumptions here (dataset_schema will be populated if and only if the query of a prepared statement returns a result set)? I'm wondering if there might be systems that don't analyze the query at first and always omit the schema until actual execution

Yes.

However, I think all our queries return results (InfluxDB doesn't do DDL / DML using SQL -- there is an alternate write path) so this particular issue is unlikely to affect us

Our system will actually analyze / plan the query twice -- once on the prepare and once when executing. While this is somewhat wasteful it allows us to remain stateless between requests

[ennuite]

@lidavidm Hello David. I currently work at Dremio. From our side we are ok with these changes.