Select paths individually in parallel rather than using "in clause"

[tjamesturner]

our cyanite stack performs substantially better if we request paths individually rather than in bulk. specifically, if we request 100 metrics in a single select query, it takes roughly 10 times longer than selecting those same 100 metrics one at a time. this seems to be a problem with heap space on the coordinating cassandra node. as such, i suggest we change the fetch query to retrieve one metric at a time, then sort the returned rows by time within cyanite. that way there is substantially less heap pressure on cassandra, and only slightly more cpu load on cyanite.

i have branched the latest 0.1.3 version in order to make the change. my revision can be found here: https://github.com/tjamesturner/cyanite/commit/0ba66249fd9ace7ff92266a0df02828c55e67b0f

[pyr]

Thanks @tjamesturner. Interesting approach, I'll take on this once the query API is updated.