API, Core: Introduce classes for content stats

[nastra]

This introduces just the base classes that are needed for content stats and is extracted from #13694.

[RussellSpitzer]

I think this is all pretty great, I have some minor comments but I do think before we merge we should finalize the vote on the design.

[pvary]

What do we think about the error handling? I see that in many places we just return -1.
I think we would like to highlight the cases when we have unexpected results. We might decide to keep writing out data when the stats are facing unexpected input, but I think minimally we need to log warn messages.

WDYT?

[danielcweeks]

Can't we exclude NaN if the type isn't floating point?

[nastra]

we can, but that implies that we have set the respective field statsStruct on each FieldStats instance so that we can properly fetch the right value whenever get(int pos, Class<X> javaClass) is called. This makes the impl more complicated and would technically require to set that statsStruct everywhere where a FieldStats instance is created.

This means that

BaseFieldStats<?> fieldStatsOne = BaseFieldStats.builder().fieldId(1).build();
BaseFieldStats<?> fieldStatsTwo = BaseFieldStats.builder().fieldId(2).build();
BaseContentStats stats =
        BaseContentStats.builder()
            .withTableSchema(...)
            .withFieldStats(fieldStatsOne)
            .withFieldStats(fieldStatsTwo)
            .build();

would now become

BaseFieldStats<?> fieldStatsOne = BaseFieldStats.builder().fieldId(1).statsStruct(...).build();
BaseFieldStats<?> fieldStatsTwo = BaseFieldStats.builder().fieldId(2).statsStruct(...).build();
BaseContentStats stats =
        BaseContentStats.builder()
            .withTableSchema(...)
            .withFieldStats(fieldStatsOne)
            .withFieldStats(fieldStatsTwo)
            .build();

This makes building a new FieldStats instance quite complicated. Another alternative would be to pass the statsStruct whenever the ContentStats instance is read. I did a quick POC and that would look something like this: b676e1c

[danielcweeks]

We should add comments since this is confusing and related to the metadata file field reserved space.

[nastra]

I've slightly renamed the constants and also added comments everywhere

[danielcweeks]

Thanks @nastra !

[nastra]

thanks for the reviews @pvary, @stevenzwu, @danielcweeks