Set workspace to null for remote datasets

[etj]

Follow-up to #14524. store is already null for remote datasets. workspace cannot follow the same way yet, because unlike store it is read directly (not only through is_remote_resource()-guarded GeoFence paths) in several places that assume a non-null value.

Line references below are pinned to 246ecde5 on branch 14524_remove.

What is already safe

Every GeoFence-related caller of get_dataset_workspace() is covered by the #14524 fixes:

• geoserver/manager.py's
  remove_permissions/set_permissions read it, but are only reached through the dispatcher in resource/manager.py, which skips remote resources.
• geoserver/security.py's sync_resources_with_guardian()
  skips remote resources before it gets there.
• delete_geofence_rules_for_layer / allow_layer_to_all have no production callers.

What blocks nulling it

1. ArcGIS harvester reuses the same workspace variable for both the field value and the alternate string.
See arcgis.py lines 208-209 and 233
(the same pattern repeats at lines 312-313 and 337):
workspace = "remoteWorkspace" feeds both alternate = f"{workspace}:{name}" and "workspace": workspace in the same dict.
Nulling workspace naively turns alternate into "None:<name>".
Needs decoupling: build alternate from a literal, set additional_parameters["workspace"] = None separately.

2. set_style has no remote guard.
resource/manager.py:480
calls self._concrete_resource_manager.set_style(...) unconditionally (gated only on sld_file/sld_uploaded being present on an update call), with no is_remote_resource() check. That reaches geoserver/helpers.py's set_dataset_style() / save_style(), which call
gs_catalog.get_style(name, workspace=saved_dataset.workspace) / create_style(..., workspace=...) against the local GeoServer.

Today workspace="remoteWorkspace" there is harmless: GeoServer has no such workspace, so the lookup just misses. With workspace=None, get_style()/create_style() mean unscoped/global style, which is a real behavior change and could collide with an unrelated global style of the same name.
This gap predates #14524 and should be fixed regardless of the workspace change.

3. Persisted, user-visible link names embed the raw value.
utils.py lines 1530, 1545, 1566:
ogc_wms_name = f"OGC WMS: {instance.workspace} Service" (and the WFS/WCS
equivalents). With workspace=None these become literally "OGC WMS: None Service" on every remote dataset's page. Not a crash, but a visible regression.

4. Dataset.workspace is not nullable today.
layers/models.py:120:
workspace = models.CharField(_("Workspace"), max_length=255), no null=True.
A migration is needed, same shape as layers/migrations/0047_alter_dataset_store.py.

Tasks in scope

1. Migration: make workspace nullable.
2. Harvesters (wms.py, arcgis.py x2) and service handlers set workspace: None in their additional_parameters / indexed-dataset fields, decoupling ArcGIS's alternate construction from the field value.
3. Guard resource/manager.py's set_style dispatch with is_remote_resource().
4. Fix the three OGC link-name f-strings in utils.py to omit or fall back sensibly when workspace is falsy.