Improve experience when `dualPublish` is configured

The following 2 problems exist when `dualPublish` is configured:

- If the secondary publish succeeds, its build scan is written to the `build-scans.csv` file and its build scans are used to create the summary
- If the secondary publish fails but the primary publish succeeds, the experiment aborts (see below)


We should consider making the following changes:

- If the secondary publish succeeds, do not write its contents to `build-scans.csv` and always use the build scans from the primary publish to create the summary
- If the secondary publish fails but the primary publish succeeds, do not abort the experiment and continue

We need to confirm that the callbacks from the primary publish will always complete before the callbacks of the `dualPublish`.