Properly handle initialization failures inside the VM

[Omrigan]

Problem description / Motivation

To initialize the runtime we have a bunch of scripts triggered by busybox init or by compute_ctl. Failures in the initialization might go unnoticed and trigger other initialization steps to fail.

This issue is a follow-up for this investigation: https://neondb.slack.com/archives/C06SJG60FRB/p1727851630611959

TLDR: an incompatible tunefs binary failed as a part of the mounts.sh; script terminated early, aborting next mounts steps. This went unnoticed, until the compute failed on a workload.

DoD

1. Report the failures through alerts/early exit.
2. Make steps as independent as possible - failure in one feature should not fail other features.

Implementation ideas

We can use neonvm-daemon to do all runtime bootstraping, thus controlling every step and reporting failures/failing healthcheck.