Fix outputoffset when build dag req

[zhexuany]

Say we have a query select id, id, it, it from t.
In the past, we construct a DAG req as follows:

table scan 
    addCol
        id, id, it, it
        0    1,  2 , 3
DAGReqBuilder
    addOutputOffset 
        0 1 2 3

It works in old TiKV, but not in the latest tikv.

The expected table scan is as follow:

table scan 
    addCol
        id, it
        0    1
DAGReqBuilder
    addOutputOffset 
        0 0 1 1

This PR fixes this.

[zhexuany]

/run-all-tests

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@ce6e612). Click here to learn what that means.
The diff coverage is 75%.

@@           Coverage Diff            @@
##             master    #615   +/-   ##
========================================
  Coverage          ?   53.9%           
  Complexity        ?     983           
========================================
  Files             ?     136           
  Lines             ?    6402           
  Branches          ?     759           
========================================
  Hits              ?    3451           
  Misses            ?    2626           
  Partials          ?     325

Impacted Files	Coverage Δ	Complexity Δ
...cap/tikv/operation/iterator/CoprocessIterator.java	56.66% <ø> (ø)	5 <0> (?)
.../main/java/com/pingcap/tikv/meta/TiDAGRequest.java	41.1% <75%> (ø)	46 <0> (?)

---

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 ce6e612...1f8ca81. Read the comment docs.

[marsishandsome]

it will be better if bug fixing & refactor are in separated PR

[zhexuany]

/run-all-tests tikv=release-2.1 tidb=release-2.1 pd=release-2.1

[marsishandsome]

LGTM

[zhexuany]

@marsishandsome PTAL. In this PR, only fix will be merged into master.

[zhexuany]

/run-all-tests

[birdstorm]

LGTM computeIfAbsent would be better, but we can leave it to next PR.

[marsishandsome]

LGTM