Support stateful operators for streaming

[shuai-xu]

Hi community,
We are now working on supporting flink in gluten, and we use velox as the native engine. However, velox's operators are mainly for batch processing, it has many differences with flink. So we temporarily add some stateful operators in our own repository. We added them in a seperated directory under velox/experimental/stateful/. Now we have made them successfully working for the nexmark benchmark of flink, most frequently used operators such as Window, Aggregator, Rank, GroupBy have been supported with state.
There are many teams or companies are doing the same work(such as disscussion ##4389), most of them choose velox as the backend as velox is aimed for supporting all kind of computing contains batch, interactive and streaming.
In the future, we want to contribute them back to velox community. I'd like to know, whether the velox community is instrested in it? Is it better to add stateful releated operators as a seperated feature or integrate them with the existing operators? Looking forward for your feedback, thank you.

[kgpai]

cc: @pedroerp @mbasmanova

[mbasmanova]

@shuai-xu This is very interesting. Thank you for reaching out. We'd be interested in extending the collection of existing (batch) operators with streaming operators. I would think it makes sense to keep these operators separately though. Would you like to open a few PRs to start the process?

[shuai-xu]

OK, I will try to break the works into serval PRs, and submit them later.

[pedroerp]

+1. Stateful operators for streaming aggregations (and similar) are indeed something in scope for Velox and a recurrent ask from the community. It would be great if you were willing to contribute them. Like you pointed out, perhaps they could go under velox/experimental/streaming initially?

Cc: @xiaoxmeng @kunigami

[shuai-xu]

@pedroerp @mbasmanova I have submit a pr #14873 for the base framework to support streaming, please help to review it. Thank you very much.