Fix cases where KMetadata interface has nil object

[serathius]

DO NOT MERGE, this is one of the alternatives considered to fix kubernetes/kubernetes#100155

What this PR does / why we need it:

Fixes passing nil object to klog.KObj

Which issue(s) this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close the issue(s) when PR gets merged):
Fixes kubernetes/kubernetes#100155

Special notes for your reviewer:

Added benchmark run them with go test -bench=BenchmarkKObj -count=100 -benchmem -run=^$ and got results analysed by benchstat:

name    old time/op    new time/op    delta
KObj-8    7.04ns ± 6%    7.50ns ± 8%  +6.57%  (p=0.000 n=89+93)

name    old alloc/op   new alloc/op   delta
KObj-8     0.00B          0.00B         ~     (all equal)

name    old allocs/op  new allocs/op  delta
KObj-8      0.00           0.00         ~     (all equal)

We see around 19% worse performance by introducing reflections, still looks acceptable.

Release note:

NONE

[serathius]

Removing WIP as benchmark results look reasonable

[serathius]

Fixed how benchmark is implemented

[dims]

LGTM, i'd like a second opinion/review from @liggitt or @DirectXMan12

[serathius]

Updated benchmark results, looks better around 7% difference.

[dims]

/approve

leaving /lgtm for @liggitt or @ehashman

i can cut a release once this is merged

[serathius]

done, PTAL @liggitt