feat: Split statistics components computation & output

[MelReyCG]

Issue: #3822
On top of PR : #3829

Planned refactor: The goal is to have a way to query statistics for solver (reservoir reference pressure for wells), with a clear separation with output statistics component.

The solution proposed here aims at giving a single data storage for each component needing statistics computation. In practices, the statistics are no longer recomputed for each convergence step, but only once, at start of each iteration (it gives the same result since the operation is read only).

To achieve these goals, the PR propose to separate each roles, instead of having everything in each *PhaseStatistics class:

• RegionStatistics classes for resulting data storages at different levels:
  • sub-regions mpi-local kernel compute read-backs,
  • (cell element) regions,
  • solver discretisation MeshLevel,
• StatsAggregator classes to compute & reduce the statistics to rank 0, can be instanciated as solver helpers (as done for wells in the PR),
• StatsTask Group, the user component to schedule flow statistics computation & output to log & CSV.

Old stats data architecture (stats are sticked to regions, and shared by all components):

Problem : ProblemManager
|-> domain : DomainPartition
    |-> MeshBodies : Group
        |-> cartesianMesh : MeshBody
            |-> meshLevels : Group
                |-> Level0 : MeshLevel
                |   |-> nodeManager : NodeManager
                |   |   |  * RegionStatistics (aggregate, mpi reduced)  <- stats data
                |   |   |-> sets : Group
                |   |       | * all : Wrapper< index array >
                |   |       | * xneg : Wrapper< index array >
                |   |       [...] (other element sets)
                |   |
                |   |-> ElementRegions : ElementRegionManager
                |   |   |  * RegionStatistics (aggregate, mpi reduced)  <- stats data
                |   |   |-> Channel : CellElementRegion
                |   |   |   |-> cb-0_0_0 : CellElementSubRegion
                |   |   |   |   | * pressure : Wrapper< real64 array >
                |   |   |   |   | * temperature : Wrapper< real64 array >
                |   |   |   |   | * RegionStatistics (compute read-back)  <- stats data
                |   |   |   |   [...] (other fields)
                |   |   |   |
                |   |   |   |-> cb-0_0_1 : CellElementSubRegion
                |   |   |   |   | * pressure : Wrapper< real64 array >
                |   |   |   |   | * temperature : Wrapper< real64 array >
                |   |   |   |   | * RegionStatistics (compute read-back)  <- stats data
                |   |   |   |   [...] (other fields)
                |   |   |   |
                |   |   |   [...] (other sub-regions)
                |   |   |
                |   |   |-> Barrier : CellElementRegion
                |   |       |-> cb-1_0_0 : CellElementSubRegion
                |   |       |-> cb-1_0_1 : CellElementSubRegion
                |   |       |-> * RegionStatistics (compute read-back)  <- stats data
                |   |       [...] (other sub-regions)
                |   |
                |   [...] (other element managers)
                |
                [...] (other discretizations)

Proposed new way, with a clear storage for each aggregator instance:

Problem : ProblemManager
|-> domain : DomainPartition
    |-> MeshBodies : Group
        |-> cartesianMesh : MeshBody
            |-> meshLevels : Group
                |-> Level0 : MeshLevel
                |   |-> nodeManager : NodeManager
                |   |   |-> sets : Group
                |   |       | * all : Wrapper< index array >
                |   |       | * xneg : Wrapper< index array >
                |   |       [...] (other element sets)
                |   |
                |   |-> ElementRegions : ElementRegionManager
                |   |   |-> Channel : CellElementRegion
                |   |   |   |-> cb-0_0_0 : CellElementSubRegion
                |   |   |   |   | * pressure : Wrapper< real64 array >
                |   |   |   |   | * temperature : Wrapper< real64 array >
                |   |   |   |   [...] (other fields)
                |   |   |   |
                |   |   |   |-> cb-0_0_1 : CellElementSubRegion
                |   |   |   |   | * pressure : Wrapper< real64 array >
                |   |   |   |   | * temperature : Wrapper< real64 array >
                |   |   |   |   [...] (other fields)
                |   |   |   |
                |   |   |   [...] (other sub-regions)
                |   |   |
                |   |   |-> Barrier : CellElementRegion
                |   |       |-> cb-1_0_0 : CellElementSubRegion
                |   |       |-> cb-1_0_1 : CellElementSubRegion
                |   |       [...] (other sub-regions)
                |   |
                |   [...] (other element managers)
         ____   |   |
         |      |   |-> statistics : Group (storage for all stats)
         |      |       |-> flowStats : Group (storage for this instance stats)
         |      |       |   |-> regionsStatistics : RegionStatistics (aggregate)
         |      |       |       |-> Channel : RegionStatistics (aggregate, mpi reduced)
         |      |       |       |   |-> cb-0_0_0 : RegionStatistics (compute read-back)
 stats   |      |       |       |   |-> cb-0_0_1 : RegionStatistics (compute read-back)
 data -> |      |       |       |   [...] (other sub-regions stats)
         |      |       |       |
         |      |       |       |-> Barrier : RegionStatistics (aggregate, mpi reduced)
         |      |       |           |-> cb-1_0_0 : RegionStatistics (compute read-back)
         |      |       |           |-> cb-1_0_1 : RegionStatistics (compute read-back)
         |      |       |           [...] (other sub-regions stats)
         |      |       |
         |___   |       [...] (other stats storages)
                |
                [...] (other discretizations)

Decoupling the storage of the CellElementRegion will help to provide a way of obtaining statistics from arbitrary mesh regions (defined by sets for instance).

---

PR also propose to consolidate testFlowStatistics by also checking the total mass balance of the compositional cases.

[tjb-ltk]

What is a PreparedTableLayout vs presumably an "non" PreparedTableLayout

[MelReyCG]

The TableLayout is the descriptor of the desired layout, and the PreparedTableLayout has precomputed data for optimizing further ascii formatting.

[tjb-ltk]

Why is this in the getTotalFluidMass function and not inside the region stats... could we get rid of the specializations at the getter level...