Issues left by cohort processor refactoring

[Zrealshadow]

Create an issue and submit a corresponding PR for each one.

1. need to add the logic to check the meta chunk to speed up the query processing.
2. remove aggregation logic in ValueSelection.
   • ValueSelection only handles filtering (can be merged with EventSelection)
   • RetUnit will encapsulate the logic of updating statistics (discussed below)
3. Factory class to handle FieldRS creation (better encapsulation)
4. rework the ProjectedTuple
   • existing impl aims to mimic the old logic which introduces additional indirections
   • keep it simple, let its producer decides on which indexed value to retrieve from it.
   • make it immutable. avoid loadattr that mutates internal data.
   • separate handling for special field: user id and action time
5. MetaFieldRS to create value converter/translator (extensibility and polymorphism)
   • currently we only have two types, we can expect having more field types.
   • The MetaFieldRS translates the token (gid for string now) stored in data chunk to actual values. (no-op for range)
6. Augment RetUnit
   • perhaps we should rename this variable
   • it will contain additional variables for max, min, etc. (now only counts)
   • A solution is to have a list of variable and keep a list functions (aggregators) that take in new value and update its corresponding variable.
7. add documentations: DataHashFieldRS Need to add descriptions to the assumptions: all input vector ever used there have efficient implementation of getting by index (Zint, BitVector, ZintBitInput,)