[ri-cloud-api]: Add grid data from sumo using api

[jorgenherje]

Loads Eclipse grid data (roff) from Sumo as a RimReservoirGridEnsemble, one case per realization.

• Async read/fetch of grid property per timestep.
• New: RiaSumoGrid (transport), RifReaderSumoGridProperty (lazy property reader), RimReservoirGridEnsembleSumo, RimRoffCaseSumo, and a "Create Reservoir Grid Ensemble" command.
• Shared grid: Sumo reports IJK per realization, so grid sharing is decided before any download. When dimensions match, one RigMainGrid is kept for the whole ensemble; each realization still parses its own blob to read its own ACTNUM, then releases the geometry. Realizations open lazily.
• Folders: Summary/Sumo → Summary/Cloud, RimSumoDataSource → Cloud/. Folder names the concept, filename the backend. PDM keywords unchanged, so existing project files load.
• UI: cloud data sources now also in the 3D window's Data Sources dock; each window offers only its own command.
• Fixes: realizations are re-fetched on project load — without this, editing the realization filter after a load deleted every realization case. Realizations the grid has no data for are now reported instead of silently skipped.

[magnesj]

When an ensemble is created, auto-select the first grid if there are grids available. This will ensure that the creation of a grid ensemble is a one-click operation.

[magnesj]

Code review finding: prefetch-batch failures can be silently re-requested

File: ApplicationLibCode/Application/Tools/Cloud/RifReaderSumoGridProperty.cpp (requestTimeStepsAsync/onTimeStepArrived ~L224-297, timeStepsToFetch ~L400-436)

Only the displayed step (steps.front()) gets a placeholder written into its result slot via fillWithUndefinedValues (from dynamicResult). Look-ahead/look-behind steps in the same prefetch batch never get a slot placeholder until dynamicResult is actually called for them.

If one of those non-displayed prefetch steps fails, onTimeStepArrived erases it from m_pending and returns — the slot stays empty. On the next call to timeStepsToFetch (e.g. another redraw at a nearby step index, before the user navigates onto the failed one), that step's slot is still empty and it's no longer in m_pending, so it gets added back to unloadedSteps and re-requested — repeating the "flood of requests" pattern this PR intentionally fixed for the displayed step.

Impact: Not a crash, but it partially undermines the stated "stop retrying failures" fix for prefetch-batch members: repeated failed network requests can occur as the user scrubs near (not necessarily onto) a time step whose data failed to load.

Suggested fix: Write a placeholder (or track a distinct "failed" set) for every step in the prefetch batch, not just the displayed one, so timeStepsToFetch can tell "never requested" apart from "failed."

[magnesj]

Some minor changes, otherwise good to go

[jonjenssen]

Rif* files should probably be located inside FileInterface, not Application?

[magnesj]

Follow-up on comment from @jonjenssen

The Rif* file should be moved to the FileInterface folder.

GUI logic should be moved out of Rif* classes. Here is one option using callbacks:

class RifReaderSumoGridProperty : public RifReaderInterface
{
public:
    using PendingChangedCallback = std::function<void(const QString&)>;
    using TimeStepArrivedCallback = std::function<void(const QString&, size_t, bool)>;

    void setPendingChangedCallback(PendingChangedCallback cb);
    void setTimeStepArrivedCallback(TimeStepArrivedCallback cb);

private:
    void notifyPendingChanged() const;
    void notifyTimeStepArrived(const QString& propertyName, size_t stepIndex, bool ok) const;

    PendingChangedCallback m_onPendingChanged;
    TimeStepArrivedCallback m_onTimeStepArrived;
};

[magnesj]

Very good, this works as expected on my system.

[magnesj]

Fix the build errors due to missing includes on Linux