fix create_pointset for throughput mode (#589)

The columns of raw.csv file are
`algo_name,index_name,recall,throughput,latency,...`

`col 4` is latency and `col 3` is throughput.

Authors:
  - Yinzuo Jiang (https://github.com/jiangyinzuo)
  - Corey J. Nolet (https://github.com/cjnolet)

Approvers:
  - Divye Gala (https://github.com/divyegala)

URL: #589

python/cuvs_bench/cuvs_bench/run/data_export.py

@@ -256,11 +256,13 @@ def create_pointset(data, xn, yn):
         Filtered list of data points sorted by x and y metrics.
     """
     xm, ym = metrics[xn], metrics[yn]
+    y_col = 4 if yn == "latency" else 3
+
     rev_x, rev_y = (-1 if xm["worst"] < 0 else 1), (
         -1 if ym["worst"] < 0 else 1
     )
     # Sort data based on x and y metrics
-    data.sort(key=lambda t: (rev_y * t[4], rev_x * t[2]))
+    data.sort(key=lambda t: (rev_y * t[y_col], rev_x * t[2]))
     lines = []
     last_x = xm["worst"]
     comparator = (