Set up code coverage properly

[clux]

Getting it from 45% to 55% then 60% by running some of the ignored lib tests in k3d on github actions and then porting a few example basics into ignored tests. Also moved from coveralls to codecov since it seems more modern (and was easy to plumb with the action).

We should have a higher coverage than what it shows if we account for examples, but we need to either add an extra coverage run for examples (it is very slow with building all), or port some example tests into kube (like what I did for crd_api and dynamic_api here).

Ultimately, what's shown here is a tradeoff. Doc tests are slow, seems to require nightly. Example tests are slow because we need to build and run all of them sequentially. It's probably better to focus on coverage the traditional way with some core tests in kube or the relevant crates.

It doesn't seem to count kube-derive properly. Maybe it's possible to tweak the command to get that in. See config "reference"

Commands to replicate what is running here (minus output format) is:

cargo tarpaulin -- --test-threads 1

with cargo-tarpaulin 0.18.5 and the new embedded tarpaulin.toml to encapsulate flags (needed a lot of tweaks to get the ignored, feature-dependent tests in kube/src/lib to run).

For now this is at least enough to satisfy the "most code is covered" requirement from #737.

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (master@20642bb). Click here to learn what that means.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master     #740   +/-   ##
=========================================
  Coverage          ?   60.34%           
=========================================
  Files             ?       52           
  Lines             ?     3051           
  Branches          ?        0           
=========================================
  Hits              ?     1841           
  Misses            ?     1210           
  Partials          ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 20642bb...03de8ab. Read the comment docs.

[clux]

Can run --doc tests as well, which adds half a percent in coverage (because most of them are marked as no_run), but adds 5 minute to runtime on actions.

[clux]

Managed to get ignored lib tests to run using a tarpaulin.toml config file, and pushed coverage a few more percent without having to turn on nightly and doc tests (which added tons of time). Happy enough with this, but obviously things we could do to make this number a little more attractive.

[clux]

A few more percent, at a respectable 60 now. Going to leave it now, don't want to mess more until there's a base report and we see how the reports come in from the bot as PRs come in. A cool feature of this is that it doesn't make new comments, it just updates the existing comment.

[clux]

I am just going to assume that this is uncontroversial and merge this for now. If the coverage message gets annoying, we can remove the pr/push build, but need to test after merging so we have a baseline.