The start of documentation 2.0

[glwagner]

This PR reorders the documentation and will add a Grids and Fields page to the top-level of the docs. It will also clean up minor outstanding issues.

Longer description of our intent: #3672

Closes #3649

[glwagner]

@navidcy I have also enabled push previews for this PR but we can disable before merging.

[glwagner]

We'll be looking here for previews: https://clima.github.io/OceananigansDocumentation/previews/PR3673

[glwagner]

@navidcy I think it would be cool to show a visualization of the grids in the "Grids tutorial"

[francispoulin]

We'll be looking here for previews: https://clima.github.io/OceananigansDocumentation/previews/PR3673

I tried this link but received a 404 error. Does it work for others?

[glwagner]

We'll be looking here for previews: https://clima.github.io/OceananigansDocumentation/previews/PR3673

I tried this link but received a 404 error. Does it work for others?

It errors for me too! Nothing will come up until the docs build passes, looks like it is still failing.

[glwagner]

I think we should improve the docstring for Distributed:

Oceananigans.jl/src/DistributedComputations/distributed_architectures.jl

Lines 174 to 205 in 8f55656

	"""
	Distributed(child_architecture = CPU();
	communicator = MPI.COMM_WORLD,
	devices = nothing,
	synchronized_communication = false,
	partition = Partition(MPI.Comm_size(communicator)))
	Return a distributed architecture that uses MPI for communications.
	Positional arguments
	====================
	- `child_architecture`: Specifies whether the computation is performed on CPUs or GPUs.
	Default: `CPU()`.
	Keyword arguments
	=================
	- `synchronized_communication`: If `true`, always use synchronized communication through ranks.
	Default: `false`.
	- `partition`: A [`Partition`](@ref) specifying the total processors in the `x`, `y`, and `z` direction.
	Note that support for distributed `z` direction is limited; we strongly suggest
	using partition with `z = 1` kwarg.
	- `devices`: `GPU` device linked to local rank. The GPU will be assigned based on the
	local node rank as such `devices[node_rank]`. Make sure to run `--ntasks-per-node` <= `--gres=gpu`.
	If `nothing`, the devices will be assigned automatically based on the available resources
	- `communicator`: the MPI communicator, `MPI.COMM_WORLD`. This keyword argument should not be tampered with
	if not for testing or developing. Change at your own risk!
	"""

A few comments:

• Can we explain "synchronized communication" better? This is confusing --- I think it actually has to do with the algorithm a model uses (very far away from building the architecture). We have to explain that this concept is irrelevant unless we are using a model that has an asychronous algorithm. Or otherwise more specifically explain what this means.
• The entry for devices is missing a period a the end.

In general more explanation of the keyword arguments that is as local to the concept of grids as possible. The docstring makes vague references to "support for partitioning". But this refers to models, not grids. It doesn't really make sense in this context.

Also we need examples.

[navidcy]

The distributed doctests in grids.md don't run, somehow the environment fails to install?

[navidcy]

@glwagner I removed the "Distributed Grids" part from the docs. I suggest we merge this if Docs built and open a new PR with the distributed grids part of the tutorial?