Optimize FilterExec::statistics / don't ignore errors

[alamb]

Is your feature request related to a problem or challenge?

This came up from looking at the backgrace in #7522 where we saw a significant amount of planning time going to creating DataFusionError which were then thrown away.

I tracked this down to creating the error (which is ignored) here:

https://github.com/apache/arrow-datafusion/blob/c47ae0d13dbc9474897609f3dea4f80a93cf4de0/datafusion/physical-plan/src/filter.rs#L206-L209

Describe the solution you'd like

I would like less time being spent creating erorrs that are thrown away. I can see two potential ways to do this:

1. Stop ignoring the error and instead fix whatever the underlying issue is
2. Cache the calculation of the statistics so it doesn't get called multiple times

Both might be appropriate

Describe alternatives you've considered

No response

Additional context

cc @berkaysynnada as I believe this came in #6982

I don't think this is a critical fix, but I wanted to file it given I had tracked it down

[liukun4515]

I also find this thing when I do this pr #7428
The statistics can't be passed or pass an error result when meet error in the process.

I think the implementation of the filter to get statistics has some issues:
From the
https://github.com/apache/arrow-datafusion/blob/c47ae0d13dbc9474897609f3dea4f80a93cf4de0/datafusion/physical-plan/src/filter.rs#L208
https://github.com/apache/arrow-datafusion/blob/c47ae0d13dbc9474897609f3dea4f80a93cf4de0/datafusion/physical-plan/src/filter.rs#L194
when meet the error or abnormal case, just return the default statistics.

I think we should change the implementation and return the child statistics If we can't get or calculate the right statistics.

[liukun4515]

1. Stop ignoring the error and instead fix whatever the underlying issue is

I think this is the root solution.