Mount host /lib/modules for precompiled drivers on SUSE

[abrarshivani]

Description

Fixes #2776.

getDriverAdditionalConfigs builds the lib-modules volume and mount inside the
if !cr.Spec.UsePrecompiledDrivers() branch, and the block itself is guarded by
if cr.Spec.UsePrecompiledDrivers() && (pool.osRelease == "sles" || pool.osRelease == "sl-micro").
The inner condition can never hold in that branch, so an NVIDIADriver with
usePrecompiled: true on sles or sl-micro renders a driver pod without
/run/host/lib/modules.

Rather than repeat the predicate, the mount now hangs off the non-precompiled
branch as an else. The two paths are mutually exclusive, and saying so once
gives precompiled-only logic an obvious home, which is what went wrong here in
the first place. The comment above the mount goes with it: the else already
says precompiled, the guard already says SUSE, and the volume literals already
say host module tree.

This only affects the NVIDIADriver path. The equivalent code in
controllers/object_controls.go sits before the early return for precompiled
drivers and works as intended. That is also why it went unnoticed, since the
tests that came in with a7323f5 only covered the ClusterPolicy path.

One note for triage: the issue lists v24.9.2 under Environment, but v24.9.2 has
no lib-modules handling in either driver path, so the unreachable code
described in the report cannot exist there. The line numbers quoted in the issue
match current main. A v24.9.2 user on precompiled SUSE does see the same missing
mount, but because the feature had not landed yet rather than because of this
bug.

Checklist

• No secrets, sensitive information, or unrelated changes
• Lint checks passing (make lint)
• Generated assets in-sync (make validate-generated-assets)
• Go mod artifacts in-sync (make validate-modules)
• Test cases are added for new code paths

Testing

Added TestDriverPrecompiledLibModules, a table-driven test covering sles and
sl-micro with precompiled drivers, plus sles without them and ubuntu with them
as negative cases. Each case renders the DaemonSet and checks the pod spec for
the lib-modules hostPath volume and nvidia-driver-ctr for the mount, so it
covers both the helper and the template threading AdditionalConfigs into the
pod.

The two SUSE cases fail against the current code with "expected a lib-modules
volume on the driver pod spec" and pass with the fix. Removing the
AdditionalConfigs mounts from the driver container in
manifests/state-driver/0500_daemonset.yaml also fails the test, which the
helper-level assertions alone would not have caught.

go test ./internal/... ./controllers/... passes, along with go vet, gofmt
and -race. Lint and the generated-asset check were run with GOOS=linux,
since the vendored filepath-securejoin/pathrs-lite is linux-only and breaks
the typecheck of cmd/nvidia-validator on macOS. Regenerating the CRDs produces
no diff.

[tariq1890]

Thanks for working on this @abrarshivani !

This will need to be backported to release-26.7

[abrarshivani]

@tariq1890 Sure, will backport this change once this PR is checked in.

[tariq1890]

/cherry-pick release-26.7