feat: add spec.additionalVolumeMounts for the main container

[IgorOhrimenko]

What

Adds spec.additionalVolumeMounts — extra volume mounts for the Dragonfly main container. It is the counterpart to the existing spec.additionalVolumes, which adds volumes to the pod but does not mount them anywhere.

Why

additionalVolumes on its own is half a feature: you can add a volume to the pod but cannot mount it into the Dragonfly container. Pairing the two lets users compose several patterns without any snapshot-specific API, e.g.:

• Back the snapshot dir (spec.snapshot.dir) with an emptyDir. Dragonfly saves on SIGTERM and reloads on start, so an emptyDir snapshot survives in-place container restarts (e.g. a kubelet minor upgrade that changes the container hash) without pinning the pod to a node the way a local-path PVC does.
• Read-only warm-up cache — mount a golden snapshot read-only; every start loads it, runtime writes never touch the on-disk copy.
• Combine an init-seeded emptyDir for warm-up + crash protection.

Usage examples (all validated end-to-end on a cluster): https://gist.github.com/IgorOhrimenko/d679e593637e0fbe62525f165fd17d78

Implementation

• Merged into the main container by name (a user mount replaces the operator's default on collision), mirroring how additionalVolumes / additionalContainers already merge.
• Applied before the additionalContainers merge, since that merge prepends additional containers and would otherwise shift the main container off index 0.

Testing

• Unit tests in internal/resources, including one asserting mounts land on the main container only and never on additionalContainers.
• Added a config/samples example (v1alpha1_dragonfly_emptydir_snapshot.yaml).
• Verified on a live cluster: emptyDir survives simultaneous both-pod container restarts (data preserved); a read-only seed stays immutable across restarts; the init-seed + emptyDir combo warms from a baseline and keeps runtime growth across in-place restarts.

Backward compatible — the field is optional and existing specs are unaffected.

Note: I regenerated config/crd/bases (make manifests) but intentionally left the manifests/ bundle untouched — it is already out of sync with config/ on main (missing some recent upstream API-type updates), so regenerating it here would add unrelated churn. Happy to include a bundle regeneration if you prefer.

[IgorOhrimenko]

The build-operator failure looks like pre-existing e2e flakiness, not related to this change. The 3 failing specs (PVC single-replica, lifecycle connectivity, readiness-gate) don't use additionalVolumeMounts, and the new merge is a no-op when the field is unset (mergeNamedSlices(base, nil) returns the mounts unchanged), so the generated StatefulSet is identical to main for those tests. Failures are timing/infra (System should be stable … Timed out 60s, redis connection refused …, 4 flaked). The same suite intermittently fails on main too.