Fix uneven length panics

[Dieterbe]

Document concepts better and in particular introduce new concept of a "canonical series".
Now we massage the fetch and fix parameters such that a high resolution series will always look like a native lower resolution serie (one that has the same interval as what we're normalizing to).
In particular, they will now have corresponding timestamps, and the same amount of points.

fix #913 (panic in mergeSeries) - tried the repro case. can't repro anymore.
fix #1095 (mergeSeries and sumSeries) - can not repo either one with this code.
fix #874 (dupe)

[Dieterbe]

note to self, this patch was helpful to see what's going on

diff --git a/api/dataprocessor.go b/api/dataprocessor.go
index ff460aea..0ca7042a 100644
--- a/api/dataprocessor.go
+++ b/api/dataprocessor.go
@@ -9,6 +9,7 @@ import (
        "sync"
        "time"
 
+       "github.com/davecgh/go-spew/spew"
        "github.com/grafana/metrictank/api/models"
        "github.com/grafana/metrictank/consolidation"
        "github.com/grafana/metrictank/mdata"
@@ -647,6 +648,19 @@ func mergeSeries(in []models.Series) []models.Series {
        }
        merged := make([]models.Series, len(seriesByTarget))
        i := 0
+       for seg, series := range seriesByTarget {
+               fmt.Printf("mergeSeries  %v - len %d\n", seg, len(series))
+               lens := make(map[int]int)
+               for _, serie := range series {
+                       l := len(serie.Datapoints)
+                       fmt.Println("len points", l)
+                       lens[l] = l
+               }
+               if len(lens) > 1 {
+                       log.Error("mergeSeries detected merge of series of unequal length")
+                       spew.Dump(series)
+               }
+       }

[fkaleo]

LGTM

[Dieterbe]

should also fix #761