Restarting failed builds on Azure spawns 2 extra runs

### Conda-forge documentation

- [X] I could not solve my problem using the conda-forge documentation.

### Installed packages

```shell
N/A
```

### Environment info

```shell
N/A
```

### Issue

In the libmagma-feedstock, builds on the main branch often time-out because the builds take 5.5 +/- 0.5 hours to complete. The build time is quite variable; I don't know why and that's not the purpose of this issue.

The issue is that when I return and click the "Rerun failed jobs" button via the GitHub UI. Two extra runs are spawned. For example, after merging the recent update to version 2.7.2. The post-merge commit had two failed build variants, so I pushed the re-run failed jobs button.


https://dev.azure.com/conda-forge/feedstock-builds/_build?definitionId=18893&_a=summary

However, the result is that runs 20231012.2 and 20231012.3 are spawned (which is not necessary?). These later two runs are re-running all of the build variants, whereas run 20231012.1 is only rerunning the two failed variants (which is what I requested).