Make VersionMatch follow upstream + configure list semantics in watcher::Config

[clux]

Effectively makes the VersionMatch an exact duplicate of the upstream, and adds some builders here and there to compensate for the missing Any.

Have changed the watcher interface and made a new enum that maps onto VersionMatch instead, because it does not make sense to run a watcher against a pinned resource version; it will always use 0.

(at least unless/until we build pagination into it).

We could keep the enum that wraps resource version, but this is awkward; it would be the only one that bundles resourceVersion that way;

• watch api calls get it as a str (outside watchparams)
• get api calls (in a follow-up pr) will also not use the enum

so altogether we are just making it awkward for ourselves by trying to make it nice :(

Follow-up to #1162 before releasing.

Questions

Q: should we make NotOlderThan("0") a new default for ListParams? we want people to use the new-faster/less-taxing variant, although it is also slightly less consistent. I feel the people who want strong consistency guarantees should opt-in rather than people who don't want strong consistency to opt-out.

A: ListParams should follow the api. Opt-in is up to the user. This avoids breaking changes.

Q: should watcher default to a less consistent listsemantic for relists? it's much less of an issue because a few extra events will eventually get cleaned up anyway due to the nature of it being an infinite watch stream. Given the reports that people are passing on upstream complaining about our default watch semantics, it might better to provide a stronger opinion here so that we are not seen as a "taxing runtime to use"

A: Not changing defaults. While it makes sense for Controller use where things are eventually consistent and meant to deal with a few redundant reconciles anyway, it's a little more iffy for watcher. We emphasise this in the docs, in particular for Semantic. People like opting into performance a lot more than they have to debug less consistency (if that happens).

[nightkr]

I'd argue that slow is a vastly better default than incorrect. If shit's slow then you can always profile and fix that afterwards. If shit's broken then, well, it has probably been broken for a long time already by the time that you discover the problem.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (main@aed07be). Click here to learn what that means.
The diff coverage is 87.27%.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1171   +/-   ##
=======================================
  Coverage        ?   72.75%           
=======================================
  Files           ?       67           
  Lines           ?     5201           
  Branches        ?        0           
=======================================
  Hits            ?     3784           
  Misses          ?     1417           
  Partials        ?        0           

Impacted Files	Coverage Δ
kube-runtime/src/watcher.rs	43.38% <50.00%> (ø)
kube-core/src/params.rs	79.42% <85.71%> (ø)
kube-core/src/request.rs	96.27% <100.00%> (ø)

[clux]

I'd argue that slow is a vastly better default than incorrect.

It's not really about it being incorrect (as it will eventually be correct), but i guess that's how it will be perceived. Ok, will leave the ListParams default.

Are you OK with the default Semantic being re-list from cache for watchers? or would you rather keep the full quorum read default there as well.

[clux]

Ok. Have left default (strongly consistent) list semantics for the default Api, but made more opinionated defaults for watcher and edited in my reasoning for it in the main post.

Have cleaned up a bunch of builders, and shortened some tests as a result, but I am happy with this now. cc @nabokihms

[nabokihms]

From my perspective, everything seems great.

[clux]

Any other comments on this @nightkr ?
Keen to get some merges happening and this is the most public facing thing, whereas most other things are unstable features.

[nightkr]

LGTM