Expose "hints" to the planner

[ryannedolan]

#37 adds "subscription hints", which are passed to the control plane via templates. This is useful for things like kafka.numPartitions which are specific to the control plane.

For more advanced use-cases, we need the planner (i.e. PipelineRel etc) to have access to "hints" as well. This would enable users to influence the output (Flink) SQL generated for a pipeline. For example, hints could be used to dynamically enable/disable specific planner rules.

Calcite has a concept of "contexts" which might be useful here. See: https://calcite.apache.org/javadocAggregate/org/apache/calcite/plan/Context.html

[ryannedolan]

In the DDL, we should support Calcite hints (e.g. /*+OPTIONS(foo=bar)*/). This seems to require forking Calcite, or at least copy-pasting a substantial chunk of it, because Calcite's JDBC driver does not provide a mechanism for registering custom hints (or really any hints). Alternatively, we could wrap our planner in a Calcite Framework, which does support custom hints.

Flink SQL also supports Calcite hints, tho I don't think we need to leverage them here. Our hints should get passed through as connector configuration via WITH (foo = bar).