[CSI] NodePublishVolume must return error if filesystem was remounted to read-only after the error

[antonmyagkov]

Filesystem was automatically remounted to “ro” because of filesystem errors.
In general pod restart should help and “fsck“ should be run to fix fs errors but if new pod appears on the same node than NodeVolumeStage can be skipped.

Observed flow:

1. volume was remounted to “ro”
   2 pod was deleted
2. new pod started on the same node
3. NodePublishVolume was received first
4. NodeUnpublishVolume for terminated pod was received

Kubernetes doesn’t unstage pod in this case.

CSI driver should check in NodePublishVolume that access rights to filesystem is correct and return error for read-only filesystems(if read-write is expected).