fix: bypass cache queries for nodemetrics/podmetrics

[yulken]

This PR addresses the following issues:

• [BUG] Node usage not being updated in the cluster node list rancher#53151
• [BUG] Pod "Restarts" data not being updated on UI rancher#53145

Analyzing the case, I noticed that setting the flag features=ui-sql-cache=false made the metrics collection work as expected.
In summary, the thing is that NodeMetrics is not natively compatible with the watch verb, meaning its cache will never be updated after Steve starts.

root@cp:/home/rancher# kc api-resources -o wide | grep NodeMetrics
nodes                                                              metrics.k8s.io/v1beta1                    false        NodeMetrics                                 get,list   

Similarly, the "Restarts" data for the pod will also not be updated once it is cached, unless there is a modification to the pod itself.

[tomleb]

Hi @yulken thanks for looking into that. There should be a "synthetic watcher" somewhere in steve that essentially polls the data for non-watchable APIs. I vaguely remember it not working correctly but have forgotten about it. That would fix the NodeMetrics issue.

As for Pod restart column, I remember it being mentioned somewhere, not sure if it's your issue or some other issue 🤔, I'll have to check. To fix this, we're likely going to want to extract column data that we receive from the api server and store it differently. Basically, same thing as we do with the other duration columns.

We shouldn't bypass the cache since the new filters / sorting functionality depends on it.

[tomleb]

There we go, found it: rancher/rancher#52217

[yulken]

Thank you @tomleb. I'll investigate this synthetic watcher.