install-dev.sh: support ostree-based distros (Silverblue, Universal Blue)

Problem

The install-dev.sh installer does not work on ostree-based Linux distributions such as Fedora Silverblue, Universal Blue (Bazzite, Aurora, Bluefin), and Fedora CoreOS.

On these systems /usr is an immutable ostree deployment. The install_rpm_packages() function calls dnf install -y (or yum/zypper/rpm -Uvh), which fails because it cannot write to the read-only filesystem. The correct tool is rpm-ostree install, which layers the package into a new deployment that takes effect after reboot.

This also means the script's assumption that the gateway systemd unit is immediately available after package installation does not hold — start_user_gateway() runs before the new deployment is booted.

Reproduction

# On Fedora Silverblue 42 / Universal Blue
curl -LsSf https://raw.githubusercontent.com/NVIDIA/OpenShell/main/install.sh \
  | OPENSHELL_VERSION=dev sh

The openshell CLI binary gets installed (possibly via the shell profile installer path), but the gateway sub-package and its systemd user unit are not installed. Running openshell gateway start returns error: unrecognized subcommand 'start' — which is expected after PR #1221 removed gateway start, but the replacement systemd-based flow also doesn't work because the unit was never installed.

Suggested approach

1. Detect ostree in linux_package_method() — e.g. check for /run/ostree-booted or rpm-ostree availability.

2. Two possible strategies:

   • Layer via rpm-ostree install: Install the RPMs into a new deployment, warn the user that a reboot is required, and defer start_user_gateway() to a post-reboot step (perhaps a first-boot systemd unit or a message telling the user to run a registration command after reboot).

   • User-local install (no layering): Extract binaries to ~/.local/bin/, place the systemd unit in ~/.config/systemd/user/openshell-gateway.service (adjusting ExecStart), and enable it immediately. This avoids the reboot but diverges from the package-managed layout.

   The user-local approach is probably the better UX for a dev install since it doesn't require a reboot and doesn't touch the base image.

Environment

• Fedora Silverblue / Universal Blue
• install-dev.sh from main (post PR #1221)
• OPENSHELL_VERSION=dev

Start in install-dev.sh with linux_package_method(), install_rpm_packages(), and start_user_gateway(). Reproduce the installer behavior on an ostree-based distribution, then determine whether layering with rpm-ostree or a user-local install fits the dev flow. Done means package or user-local installation succeeds and the gateway systemd user unit has a documented, working post-install start path.