libnetwork: avoid restarting rootless helper during teardown

[shrevid03]

Summary

Avoid restarting or recreating the rootless network helper during teardown.

Teardown() currently uses the same getOrCreateNetns() path as Setup() and
Run(). If the rootless network namespace still exists but the helper process
has already exited, that path checks the stale helper PID and can fall through
to starting the helper again.

During an orderly systemd user-manager shutdown, starting a new helper scope can
fail because the shutdown transaction is already queued.

This change adds an explicit create/non-create mode:

• Setup() and Run() retain the existing create/recovery behavior.
• Teardown() only uses an existing network namespace and does not recreate or
  restart the rootless helper.

Related to podman-container-tools/podman#29419.

Testing

Added Test_getOrCreateNetnsNoCreateDoesNotRestartHelper.

The test creates an existing network namespace reference with a stale helper PID
and verifies that getOrCreateNetns(false) returns the existing namespace
without attempting to restart the helper.

Verified with:

• go test ./libnetwork/internal/rootlessnetns -count=1
• git diff --check

I reproduced the shutdown failure on a Fedora CoreOS 44 aarch64 Podman Machine
VM using Podman 6.0.2, Netavark 2.0.0, and Aardvark DNS 2.0.0 with a rootless
Quadlet container on a custom bridge network.

With this change, the rootless-netns teardown path no longer attempts to
restart the dead rootless network helper.

Scope

While validating the full shutdown path, I also identified a separate
Netavark/Aardvark teardown path that can attempt to restart a dead
aardvark-dns process.

That behavior is outside the scope of this PR; this change is intentionally
limited to the rootless-netns teardown behavior in common/libnetwork.