Introduce index:dcount() function

[Korablev77]

Currently one of mainstream algorithms is HyperLogLog which makes it possible to approximately count the number of distinct elements using almost no memory overhead. I suggest to implement index method :dcount() (d stands for distinct) which would apply HyperLogLog and return cardinality of index keys (obviously it makes sense only for non-unique indexes). I suppose it might be useful for some customers.

[unera]

I think that new index method - is wrong idea, because we have already had :count method , that could be modified:

box.space.tester.index.primary:count('Alpha!', { iterator = 'LE', distinct = true })
box.space.tester.index.primary:count('Alpha!', { distinct = true })  -- iterator = 'ALL'

[Gerold103]

You seem to misunderstand the ticket. It is not distinct count. It is approximate distinct count. Using a flag distinct would be misleading, because it won't return an entirely correct value at least sometimes.

[unera]

Ok, change disticnt=true to approximate=true, disticnt=true

My points:

1. Don't add new methods
2. Add options
3. I don't know if the index is useful for anyone.

Looks good to do if we have no other tasks.

[no1seman]

approximate, disticnt, doesn't matter... if this function has the same as Redis inplementation accurancy - seems that this will be enougth, but, why not to make this function a part of LUA API to make it possible to use it for tables as well as for indexes?