BUG: data gets saved to database twice

Nobody has claimed this yet.

### Steps to reproduce

I ran a 2D scan using the do2d function in my pull request, which is basically a copy/paste of the do2d function provided in the qcodes examples for the new dataset.

`do2d(qdac.lp, -0.75, -0.9, 500, 0.07, qdac.b, -0.5, 0.5, 100, 0.07, lockin_2.X, lockin_2.Y)`

For completeness I provide the function here:

```python
def do2d(param_set1, start1, stop1, num_points1, delay1,
param_set2, start2, stop2, num_points2, delay2,
*param_meas):
'''Scan 2D of param_set and measure param_meas.'''
meas = Measurement()
refresh_time = 1. # in s
meas.write_period = refresh_time
meas.register_parameter(param_set1)
param_set1.post_delay = delay1
meas.register_parameter(param_set2)
param_set2.post_delay = delay2
output = []
for parameter in param_meas:
meas.register_parameter(parameter, setpoints=(param_set1, param_set2))
output.append([parameter, None])
progress_bar = progressbar.ProgressBar(max_value=num_points1 * num_points2)
points_taken = 0
time.sleep(0.1)

with meas.run() as datasaver:
run_id = datasaver.run_id
last_time = time.time()
for set_point1 in np.linspace(start1, stop1, num_points1):
param_set1.set(set_point1)
for set_point2 in np.linspace(start2, stop2, num_points2):
param_set2.set(set_point2)
for i, parameter in enumerate(param_meas):
output[i][1] = parameter.get()
datasaver.add_result((param_set1, set_point1),
(param_set2, set_point2),
*output)
points_taken += 1
current_time = time.time()
if current_time - last_time >= refresh_time:
last_time = current_time
progress_bar.update(points_taken)
progress_bar.update(points_taken)
return run_id
```

### Expected behavior

Each point should get saved once.

### Actual behaviour

Some points get saved more than once (which destroys alignment of the data if your data analysis relies on it). Excerpt from the table:

...
-7,50E+17 | -2,27E+17 | 2,89E+15 | -1,14E+15
-7,50E+17 | -2,17E+17 | 2,88E+15 | -1,18E+15
-7,50E+17 | -2,07E+17 | 2,79E+15 | -1,23E+15
-7,50E+17 | -1,97E+17 | 2,74E+15 | -1,23E+15
-7,50E+17 | -1,87E+17 | 2,74E+15 | -1,19E+15
-7,50E+17 | -1,77E+17 | 2,76E+15 | -1,16E+15
-7,50E+17 | -1,67E+17 | 2,81E+15 | -1,17E+15
-7,50E+17 | -1,57E+17 | 2,86E+15 | -1,24E+15
-7,50E+17 | -1,46E+17 | 2,82E+15 | -1,30E+15
-7,50E+17 | -1,36E+17 | 2,66E+15 | -1,25E+15
-7,50E+17 | -1,26E+17 | 2,43E+15 | -1,15E+15
-7,50E+17 | -1,16E+17 | 2,29E+15 | -1,10E+15
-7,50E+17 | -2,27E+17 | 2,89E+15 | -1,14E+15
-7,50E+17 | -2,17E+17 | 2,88E+15 | -1,18E+15
-7,50E+17 | -2,07E+17 | 2,79E+15 | -1,23E+15
-7,50E+17 | -1,97E+17 | 2,74E+15 | -1,23E+15
...

In this case there is a 12 row block that gets repeated.

### System

Windows 10

**qcodes branch master**

**qcodes commit f85dfeebd02e7c9cacf925195bbd2217a45d9df7**

Start with the shown do2d entry point, especially Measurement.run() and datasaver.add_result(), and reproduce the repeated 12-row block using the supplied scan parameters. Trace how results are persisted and identify why rows are written more than once. Done means each scan point appears exactly once and the resulting data remains aligned.

Written by the indexing model from the issue text.