Add the GNKI tools and unit testing

[odunbar]

Purpose

Closes #404
Closes #405
Closes #406
Coauthored by @rgjini

Content

• Added basic GNKI struct taking in prior
• Added update function,
• Added some defaults for timestepper etc.
• Coverage of unit testing passes
• Docstrings

MISC

• Also resolves some bugs in testing where we need to deepcopy(Unscented(..)) processes when reusing them as they store a history internally
• Also resolves an "always-false" check statement where nameof(typeof(X)) == name should be Symbol(name). Replaced all cases with isa(X,name)
• Also resolves bugs in github-actions where old packages are loaded instead of new ones, and resolves a related docs issue

Example output: Sinusoid N_ens=10, N_iter=5

Ran defaults with timestep = 1
i.e.

ensemble_kalman_process = EKP.EnsembleKalmanProcess(
    initial_ensemble,
    y,
    Γ,
    GaussNewtonInversion(prior),
    localization_method = Localizers.NoLocalization(),
    accelerator = DefaultAccelerator(),
) 

and with localization + accelerator, timestep =1 "bells and whistles"

ensemble_kalman_process = EKP.EnsembleKalmanProcess(
    initial_ensemble,
    y,
    Γ,
    GaussNewtonInversion(prior),
)

---

• I have read and checked the items on the review checklist.

[rgjini]

The algorithm looks good to me!

[rgjini]

Everything looks good to me! A couple questions: there are a few places with warnings saying certain lines are not covered by the tests, is that normal? I was also curious what ekp.verbose means? Is this something all the algorithms have?

[odunbar]

Everything looks good to me! A couple questions: there are a few places with warnings saying certain lines are not covered by the tests, is that normal?

This means that we added some new lines of code that are untested - we require ~90% of the new code to be covered as a soft-rule so we can add a couple more tests to cover anything important that was missed

I was also curious what ekp.verbose means? Is this something all the algorithms have?

Yes, if we add verbose = true when creating the EKP object it prints some basic diagnostic-type information about the convergence

By the looks of it, I shall

• improve the test coverage, just by making one of the pipelines use verbose=true across all tools to increase coverage
• look at why the windows tests fail. It's nothing we did, rather it is a some package issue in the pipeline.

...Docs now fail due to different environment issue.

• resolve docs-build issue...

[odunbar]

All resolved - merging on tests passing and docs building