Remove Database::Connection.healthy?

This method was added using an EE only method, and just inverted the
result of Postgresql::ReplicationSlot. This however creates a problem: a
Connection is tied to a specific database connection, but
ReplicationSlot is tied to the main database.

When we add support for multiple databases to ReplicationSlot, we'd
somehow have to figure out what model to use for the `healthy?` class,
as there would a ReplicationSlot model per primary database. Instead of
trying to work this out, which likely would get hairy, we simply remove
the `healthy?` method in favour of using ReplicationSlot directly. It
was only used in one place anyway, ant there is no harm in using
ReplicationSlot directly in that instance.

ee/app/workers/geo/prune_event_log_worker.rb

@@ -17,7 +17,7 @@ module Geo
 
     def perform
       return if Gitlab::Database.read_only?
-      return unless Gitlab::Database.main.healthy?
+      return if Postgresql::ReplicationSlot.lag_too_great?
 
       unless ::GeoNode.secondary_nodes.any?
         Geo::PruneEventLogService.new(:all).execute

ee/lib/ee/gitlab/database/connection.rb

@@ -6,10 +6,6 @@ module EE
       module Connection
         extend ActiveSupport::Concern
 
-        def healthy?
-          !Postgresql::ReplicationSlot.lag_too_great?
-        end
-
         def geo_uncached_queries(&block)
           raise 'No block given' unless block_given?
 

ee/spec/lib/ee/gitlab/database/connection_spec.rb

@@ -7,20 +7,6 @@ RSpec.describe Gitlab::Database::Connection do
 
   let(:connection) { described_class.new }
 
-  describe '#healthy?' do
-    it 'returns true when replication lag is not too great' do
-      allow(Postgresql::ReplicationSlot).to receive(:lag_too_great?).and_return(false)
-
-      expect(connection.healthy?).to be_truthy
-    end
-
-    it 'returns false when replication lag is too great' do
-      allow(Postgresql::ReplicationSlot).to receive(:lag_too_great?).and_return(true)
-
-      expect(connection.healthy?).to be_falsey
-    end
-  end
-
   describe '#geo_uncached_queries' do
     context 'when no block is given' do
       it 'raises error' do

ee/spec/workers/geo/prune_event_log_worker_spec.rb

@@ -29,8 +29,11 @@ RSpec.describe Geo::PruneEventLogWorker, :geo do
       end
 
       it 'does nothing when database is not feeling healthy' do
-        allow(Gitlab::Database.main).to receive(:healthy?).and_return(false)
+        allow(Postgresql::ReplicationSlot)
+          .to receive(:lag_too_great?)
+          .and_return(true)
 
+        expect(GeoNode).not_to receive(:secondary_nodes)
         expect(Geo::PruneEventLogService).not_to receive(:new)
 
         worker.perform