Reduce example count in Geo registries update mutation spec

What does this MR do and why?

Mutations::Geo::Registries::Update is tested through with geo registries shared context, which parameterizes the whole spec over the 43 Geo registry types (Geo::AbuseReportUploadRegistry, Geo::UploadRegistry, ... Geo::VulnerabilityRemediationUploadRegistry). Two groups of examples in that loop don't depend on which registry type is under test:

• The class-level authorization check — specify { expect(described_class).to require_graphql_authorizations(:read_geo_registry) }. This asserts something about described_class itself, not about any per-row registry.
• The "licensed feature not available" example. Every registry model under ee/app/models/geo/*_registry.rb (all 43, verified by grep) includes Geo::ReplicableRegistry, whose declarative_policy_class always resolves to 'Geo::RegistryPolicy' (ee/app/models/concerns/geo/replicable_registry.rb:43-45). That policy's only rule is rule { can_read_all_geo }.enable :read_geo_registry (ee/app/policies/geo/registry_policy.rb), which depends solely on the current user's can?(:read_all_geo, :global) — never on the registry class or instance.
• The granular-token authorization examples. The mutation declares authorize_granular_token permissions: :update_geo_registry, boundary: :instance, ... (ee/app/graphql/mutations/geo/registries/update.rb:12-13). The :instance boundary means the check is instance-wide and never inspects the registry_id argument, so it can't vary by registry type either.

I hoisted these into one context 'with a single representative registry type', using :geo_upload_registry as the representative factory, outside the with_them loop. The resync/reverify/maintenance-mode examples stayed inside the loop, since Geo::RegistryUpdateService does dispatch differently per registry class and that coverage is real.

No coverage was removed: authorization is still exercised (once, instead of 43 times), and the per-registry resync/reverify/maintenance-mode behavior is still exercised once per registry type, exactly as before.

Spec Run Time Optimization Results

Before optimization:

• 559 examples
• Run time: 2m 34.4s

After optimization:

• 265 examples
• Run time: 1m 28.8s

Improvement:

• Example count reduced by: 52.6% (559 to 265)
• Run time reduced by: 1m 5.6s (65.6s)

Verified with:

bundle exec rspec ee/spec/requests/api/graphql/mutations/geo/registries/update_spec.rb --dry-run  # example count
bundle exec rspec ee/spec/requests/api/graphql/mutations/geo/registries/update_spec.rb            # full run

Both runs measured in the same local checkout for a fair comparison (same warm Rails boot, same DB/Gitaly state), swapping only this file's content between the pre- and post-optimization version. 0 failures on both; pending count went from 43 to 1 (the now-single representative case for denies access under when namespace enforces granular tokens, an unrelated pre-existing pending state — # namespace has no top-level group).

bundle exec rubocop ee/spec/requests/api/graphql/mutations/geo/registries/update_spec.rb reports no offenses.

Estimated CI Cost Savings

Using the same methodology as !248403 (merged): 65.6s saved per pipeline × $0.03/minute (large runners: 3x cost factor on $10/1,000 base minutes).

Scenario	Pipelines/month	Time saved/month	Savings/year
Pessimistic	15,000	273.3h	$5,904.00
Moderate	30,000	546.7h	$11,808.00
Optimistic	50,000	911.1h	$19,680.00

How to set up and validate locally

1. Checkout this branch.
2. Run bundle exec rspec ee/spec/requests/api/graphql/mutations/geo/registries/update_spec.rb.
3. Confirm 265 examples, 0 failures.

MR acceptance checklist

Evaluate this MR against the MR acceptance checklist.