nrt: overreserve: drop custom indexer, use lister

The overreserve cache needs to periodically list all the pods
on a node to perform the synchronization check.

When the cache was introduced in 0f4ed7a , adding a thin layer
over a indexer and manage a map of node -> slice of pod identifiers
seemed like simple and nice. Speed was not really a concern at time,
even though it was thought it was equal or better than trivial listing.

However we quickly found out both in field-testing and in review
that there are a bunch of corner cases to handle to manage this "thin
layer cache" and the code started to grow complex.

Scale testing of the overrserve cache is demonstrating more and more
edge cases. The "thin layer cache" is quickly losing appeal as the
complexity (projected or implemented) quickly grows.

So, we re-evaluate the whole idea and we drop the not-thin-anymore
cache and we just use the simplest solution: in the resync code,
we just use the battle-tested client-go lister, with trivial filtering
over the pod properties to check if we need to consider on a node or
not.

Performance (granted this is actually a concern, because there's
no profiling evidence this was ever a hotspot) is not a concern yet,
and the new code is way simpler and easier to reason about, so
this is a big improvement for the overreserve cache already.

Signed-off-by: Francesco Romani <[email protected]>

Author: ffromani