Getdata use pointslicepool and getdata return to pool

[Dieterbe]

PR on top of #1921 - review that first please

I will rebase this soon to make it less confusing.
This PR does two main things:

• on the query node if errors happen during getTargets() (whether remote or local or both), clean up (add to pool) any already created series, instead of discarding them
• on the read node in the /getdata path (that receives queries from query nodes for data) return slices to the pool, always (upon errors or in the no error case, after writing the response) . fix pointSlicePool issues #962 issue 1 - the suspicion is that this should significantly reduce memory usage on read nodes, at least in certain cases.

I have not yet been able to demonstrate memory savings.
Note that snapshots functionality is not working right now, but i have 2 screenshots of read nodes subjected to the same workload. memory is equivalent, but on the "after", allocation rate is significantly reduced. also notice more slices going into the pool, and all gets out of the pool succeeding.

PRE

POST

[Dieterbe]

OK, found something fairly obvious.
the memory amounts are so similar because the "min amount of memory used" is about the same for them (pool cleared)
. but both will always let their heap double due to the GOGC variable. it's just that in the 2nd case that takes more time.
So we can interpret as with this code applied, we can reduce GOGC and end up with a lower heap with only the same cost (frequency of GC's) as before. i will do that experiment next (edit: confirmed, can reduce memory by reducing GOGC and still having similar amount of GC runs as before)

PRE

POST

[shanson7]

LGTM