Rework searchers to execute on a single reader

[prateek]

The m3db index is composed of time based index blocks, each index block is a complete index for a particular time range. Each index block is composed of segments (from writes and/or bootstraps). m3ninx's searcher implementation assumed these segments did not share document ID ranges (by ID, I mean postings.ID; not []byte). This guarantee isn't possible with m3db, because we accept writes before we bootstrap (or after, in the case of topology changes). This PR relaxes the restriction -- it makes the query searchers operate on a segment at a time, instead of all the segments at once. This will work out better for query optimisation too.

Fixes one of the issues in #865

[codecov]

Codecov Report

Merging #883 into master will increase coverage by 22.85%.
The diff coverage is 86.66%.

@@             Coverage Diff             @@
##           master     #883       +/-   ##
===========================================
+ Coverage   55.58%   78.44%   +22.85%     
===========================================
  Files         392      393        +1     
  Lines       33387    33304       -83     
===========================================
+ Hits        18558    26124     +7566     
+ Misses      13164     5386     -7778     
- Partials     1665     1794      +129

Flag	Coverage Δ
#dbnode	81.35% <ø> (+11.86%)	⬆️
#m3ninx	72.39% <86.66%> (+50.96%)	⬆️
#query	67.96% <ø> (+59.62%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a852262...b32f26b. Read the comment docs.

[jeromefroe]

LGTM, makes the searcher code a lot simpler, I wonder if the fact that doc ID's can be in multiple segments will cause difficulties in other areas. Either way though this makes the searcher code better.