Field major refactor: one Field to rule them all

[glwagner]

This PR resolves #2052 by consolidating Field to encompass:

• Field
• ComputedField
• ReducedField
• AveragedField

In addition, it gets rid of KernelComputedField since that functionality is now covered by KernelComputedOperation + Field.

Field will no longer explicitly keep track of architecture, because architecture belongs to grid now.

This PR also change the API for constructing fields:

a = Field{Face, Face, Center}(grid) # constructs a field at `(Face, Face, Center)`
b = Field{Nothing, Face, Center}(grid) # constructs a field that is reduced in the x-direction at `(Face, Center)` in `y, z`
c = Field{Center, Center, Center}(grid)
csq = Field(c^2) # constructs a "computed field" that calculates and stores c^2 via `compute!(c_sq)`

There are also convenience constructors that take a tuple of locations:

loc = (Face, Face, Center)
a = Field(loc, grid)

AveragedField functionality is now wrapped into Field. Rather than writing c_sq_avg = AveragedField(c^2, dims=(1, 2)), we now write

c_sq_avg = Field(Average(c^2, dims=(1, 2)))

There is also an interface for generic reductions, so another possibility is

c_sq_max = Field(Reduction(maximum!, c^2, dims=(1, 2)))
compute!(c_sq_max)

It was already possible to compute maximum! with a ReducedField and an AbstractField; this interface simply provides a way to compute these things on the fly for diagnostics purposes while running simulations.

In summary, this PR is meant to reduce the amount of code we have to maintain, significantly declutter our Field implementations, clean up the API, and hopefully make it easier to extend Field implementations in the future, especially regarding reductions of operations such as SummedField, SlicedField, etc.

[francispoulin]

This sounds like a great idea, and a big step forward.

One question: Considering that we use Flat in Grid to specify when a dimension is not to be built, could we use Flat instead of Nothing to signify that in this PR?

[glwagner]

This sounds like a great idea, and a big step forward.

One question: Considering that we use Flat in Grid to specify when a dimension is not to be built, could we use Flat instead of Nothing to signify that in this PR?

Well, we distinguish between topologies (Periodic, Bounded, Flat) and locations (Center, Face, Nothing).

Are you suggesting that we introduce a new location called Flat, the same name as the topology Flat?

[francispoulin]

No, sorry, my silly mistake. Nothing makes a lot of sense. Please ignore my previous comment.

[glwagner]

No, sorry, my silly mistake. Nothing makes a lot of sense. Please ignore my previous comment.

It still might make sense to come up with a better location than Nothing... (I guess the meaning is that in a reduced or Flat direction, a field has "no location").

[francispoulin]

How about None? A bit shorter than NoLocation.