jepsen: (re-)enable the big-skew nemeses

[knz]

Following up to #15717.

These nemeses are currently disabled because of assumptions about the naming of network interfaces.
This needs to be solved, e.g. by adding the appropriate command flags to align the jepsen expectations with the interfaces proposed by the spawned VMs.

[bdarnell]

@knz Do you remember why we have so many different variants of the skew nemesis? We have small-skews, subcritical-skews, and critical-skews, which just mess with the clock (so they don't have network interface dependencies), and big-skews and huge-skews, which adjust the clock and inject an equivalent amount of network latency.

It seems to me that we should at least be running subcritical-skews. The big-skews nemesis creates a larger offset than we allow, so I would expect it to fail in the absence of the network manipulation. Are the big and huge skew nemeses meant to show that high network latency somehow increases our tolerance for clock offsets?

(Also, all the numbers here need to be doubled to match our increased default max-offset).

[knz]

so I wrote small-skews and big-skews initially. Both were implemented to check that our safeguards are indeed effective in the "common" case.

Kyle then added the other three; subcriticial-skews and critical-skews to check edge behaviors, and huge-skews to add the check of our HLC logic on top of what big-skews was meant to test.

I think we should deprecate big-skews in favor of huge-skews with a random factor on the network latency; I also think our prod/test clusters are already exercising small-skews, given the large observed drifts in practice. For the other two, I don't have an opinion.

[bdarnell]

My question was mainly about why some (but not all) of the skews are accompanied by network offsets (and why it's a single nemesis instead of separate network-latency and clock-offset nemeses that can be run together). But it looks like that was kyle's change. I'll turn on the subcritical skews and ignore big-skews for now.

[bdarnell]

We're now running with subcritical skews. I don't think we really need to enable big skews or skews with network delays, so I'm closing this issue.