Release process: bring back error detection

Follow-up on !1477 (merged)

The pipeline added in 75e1e64b messed up the error detection, so the hash file will be created even on failure, and then I can see how the RM could keep going and only realize later that something went wrong.

Minimal reproducer:

(
set -eu
false | tee /tmp/bla
) && echo success

→ prints "success" which is not what we want.

I've verified that adding "set -o pipefail" fixes that both in ZSH and Bash.