Make "publish binary" pipeline more resilient

[fweikert]

Currently the pipeline only executes the upload step when builds on all platforms have succeeded. This leads to two problems:

1. A failing build on a single platform prevents the upload of binaries for all other platforms even though they were built without problems.
2. A worker outage for a single platform stalls the entire pipeline forever, without sending us an alert (since the pipeline is still running and not in the "failed" state).

Consequently, we should implement two improvements:

1. Implement a hard time-out for each build step, i.e. failing them if they take too long.
2. Upload all available binaries as soon as all build steps have either passed or failed. This logic must handle retries.