[FEA] user defined hints to better optimize query execution

[binmahone]

The idea origins from a discussion in #12122 (comment) , @revans2 @mattahrens and I had some offline discussions are we decide to open a issue to track this idea first.

apache/spark#25071 (there's an example in the PR) enables us to create user defined hints to optimze query execution. Like #12121, given historical exeuction eventlogs, we may come up with some useful hints to inject into the query, either by manually or by help of some kind of automatic tools. With these hints rapids plugin can be smarter in choosing different operator implementation, e.g. Velox Scan vs. Gpu Scan, or deciding whether to do join reordering or aggregation push down.

Possible hint candidates include:

1. if a join explodes
2. if a join does not add rows
3. if a join is highly selective
4. if a scan is higly selective after predicate filter pushdown
5. etc.

Compared with #12121, hints might be able to support more fine grained optimizations. For example, in a single query one can specify different hints for different tables, so that some of the Scans can use Velox Scan and others using GPU Scan. However, 12121 and hints should NOT be mutually exclusive, as they cover slightly different use cases.

[binmahone]

@revans2 please feel free to refine the issue description