Fix manager CT startup in local Docker setup

[ArshSSandhu]

Summary

This PR fixes a local startup issue where the proxmox service can become unhealthy because Manager CT 100 does not boot correctly in Docker Desktop + WSL2 environments.

During debugging, Manager CT 100 was created with cores: 4, which caused LXC to generate an invalid empty cpuset line:

lxc.cgroup.cpuset.cpus =

That prevented CT 100 from starting. After removing the cores setting, CT 100 could start, but it still had a temporary emergency boot entrypoint configured:

entrypoint: /sbin/init systemd.unit=emergency.target

This PR makes the core limit optional and removes the temporary emergency entrypoint before the final Manager CT startup.

Changes

• Makes the Manager CT core limit optional using MANAGER_CORES.
• Removes the hardcoded --cores=4 from the default local setup.
• Removes the temporary emergency entrypoint before the final CT startup so Manager boots normally with networking and services enabled.

Testing

Tested locally on Windows + Ubuntu WSL2 + Docker Desktop.

Started from a clean Docker state:

docker compose down -v
docker compose build proxmox pull-image
docker compose up -d

Verified the Proxmox healthcheck became healthy:

docker inspect opensource-server-proxmox-1 --format '{{.State.Health.Status}}'

Result:

healthy

Also verified inside the Proxmox container:

lxc-info -n 100
grep -n "cpuset" /var/lib/lxc/100/config || echo "no cpuset line"
grep -nE "cores|entrypoint" /etc/pve/lxc/100.conf || echo "no cores or entrypoint in CT config"
pct exec 100 -- ip a
pct exec 100 -- ss -lntp
curl -kI --connect-timeout 5 https://10.254.0.2

Confirmed:

• CT 100 is running.
• No invalid empty cpuset line is generated.
• Manager receives IP 10.254.0.2.
• nginx listens on 80/443.
• Manager app listens on 3000.
• curl returns HTTP/2 200.

Notes

On the first startup, the healthcheck may briefly fail while Manager finishes booting, but it later becomes healthy once nginx is listening on 443.

[runleveldev]

We don't need to support "MANAGER_CORE_ARGS". I'm fine with no CPU pinning by default for manager.

[runleveldev]

If the entrypoint fails to delete we need to assume the container is in a bad state and not start it. Let this fail rather than swallowing the error with || true.