Optimize zunion[store] command by avoiding the duplicate copy of dict

[lyq2333]

In the past zunion and zunionstore command, we first constructed a accumulator dict and then converted it into the final sorted set. But we can use the dict in dstobj directly to avoid the cost of constructing a new dict during converting to improve performance.

In this PR, I removed the accumulator dict and replace it with the dict in dstobj.

I made a test on zunion and zunionstore command. I first add two zsets each with 100 unique elements. Then I call zunion or zunionstore on these zsets. The result shows the performance can improve about 18%.

zunion

unstable

Summary:
  throughput summary: 6488.24 requests per second
  latency summary (msec):
          avg       min       p50       p95       p99       max
        7.253     0.976     7.207     8.087     8.335    11.039

This PR

Summary:
  throughput summary: 7651.99 requests per second
  latency summary (msec):
          avg       min       p50       p95       p99       max
        6.083     0.976     6.031     6.935     7.191    10.991

zunionstore

unstable

Summary:
  throughput summary: 7036.55 requests per second
  latency summary (msec):
          avg       min       p50       p95       p99       max
        7.015     0.992     7.943     8.543     8.767    17.887

This PR

Summary:
  throughput summary: 8353.52 requests per second
  latency summary (msec):
          avg       min       p50       p95       p99       max
        5.896     0.968     6.663     7.263     7.559    16.111

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}