[FEATURE] as_dicts for DataFrames

[goodwanghan]

Is your feature request related to a problem? Please describe.
Now we only has as_dict_iterable, but if we want to get an iterable, the execution behavior can be very different. Spark is the best example, when there are a lot of partitions (thousands of), as_dict_iterable can be very slow because it will go through partition by partition.

Describe the solution you'd like
as_dicts means getting dicts as a whole, with this semantic, we could collect all data in parallel, and then convert to dicts.