New DeviceManager interface

[avalluri]

These set of changes defines/implements new PmemDeviceManger interface, this hides the actual device management part from driver

[okartau]

excellent idea to separate device interface, allows cleaner implementation of #26

[avalluri]

@okartau Now i fixed the make test failure. Can you please review.

[okartau]

This change increases csi-sanity error count to 1 (was: zero):
Node Service
.../github.com/kubernetes-csi/csi-test/pkg/sanity/tests.go:44
NodeStageVolume
.../github.com/kubernetes-csi/csi-test/pkg/sanity/node.go:244
should fail when no volume capability is provided [It]
.../github.com/kubernetes-csi/csi-test/pkg/sanity/node.go:309

Expected
    <codes.Code>: 2
to equal
    <codes.Code>: 3

.../github.com/kubernetes-csi/csi-test/pkg/sanity/node.go:325

[okartau]

looks like we return Unknown (value 2) instead of InvalidArgument (value 3) somewhere in NodeStageVolume

[okartau]

we don't have explicit check for capabilities existence in NodeStageVolume handler, not before nor after this change. Seems it still returned correct code before via some side effect. I propose we add explicit check for capabilities existence

[okartau]

make test is OK now!

[avalluri]

looks like we return Unknown (value 2) instead of InvalidArgument (value 3) somewhere in NodeStageVolume

I think i found the issue behind this failure:

+	device, err := ns.dm.GetDevice(req.VolumeId)
+	if err != nil {
+		pmemcommon.Infof(3, ctx, "NodeStageVolume: did not find volume %s", attrs["name"])
+		return nil, err
 	}

[okartau]

now csi-sanity test passes with zero errors

[okartau]

Here, and 2 lines below, perhaps could start line with function name instead of CreateDevice: and CreateVolume. At least my initial idea with those comment was, indicate function name

[avalluri]

Ok, will fix and resubmit