Retry os_updates tasks prone to failure

[tobiasmcnulty]

Some apt tasks in the os_updates role are prone to failure on some VMs. I think the apt process from the previous task sticks around for a few seconds even after the success is received by Ansible. This seems to work around it, for example:

TASK [caktus.hosting_services.os_updates : Remove dependencies that are no longer required] ***
Thursday 09 January 2025  14:08:18 -0500 (0:00:07.784)       0:00:25.391 ****** 
<snip>
FAILED - RETRYING: [db1_testing]: Remove dependencies that are no longer required (5 retries left).
<snip>
ok: [db1_testing]

Also, we might as well make os_updates_apt_cache_valid_time configurable.