RUBY-3824 Make Session#unpin idempotent

The transactions runner triggers Session#unpin twice for the same network
error on the bulk_write code path: once from the unpin_maybe handler in
Operation::Executable#do_execute, and again from the unpin_maybe handler
in BulkWrite#execute_operation wrapping the same OpMsg execution. The
second call tried to check the connection back into the pool that had
already received it, raising "Trying to check in a connection which is
not currently checked out by this pool". This regression had been hidden
since the rubocop autocorrect on the runner (commit 8db6431) skipped
all transactions tests; it surfaced as soon as PR 3033 re-enabled them
on sharded clusters.

Guard Session#unpin to return early when there is no pinned state. The
single-document path is unaffected (only one unpin_maybe layer there).

Author: comandeo-mongo

lib/mongo/session.rb

@@ -936,6 +936,13 @@ def pin_to_connection(connection_global_id, connection: nil)
     #
     # @api private
     def unpin(connection = nil)
+      # Idempotent: if there is no pinned state to clear, do nothing. Nested
+      # unpin_maybe handlers (e.g. in BulkWrite#execute_operation wrapping an
+      # OpMsg execution that already calls unpin_maybe in its own do_execute)
+      # can call this method twice for the same error; checking the connection
+      # back into the pool a second time would raise from the pool.
+      return if @pinned_server.nil? && @pinned_connection.nil? && @pinned_connection_global_id.nil?
+
       @pinned_server = nil
       @pinned_connection_global_id = nil
       conn = connection || @pinned_connection

spec/mongo/session_spec.rb

@@ -340,4 +340,26 @@
       session.session_id.should be_a(BSON::Document)
     end
   end
+
+  describe '#unpin' do
+    context 'when called twice with the same connection' do
+      # Nested unpin_maybe handlers in the bulk_write code path can trigger
+      # Session#unpin to be invoked twice for the same network error: once
+      # from Operation::Executable#do_execute and once from
+      # BulkWrite#execute_operation. The second call must be a no-op rather
+      # than raising "Trying to check in a connection which is not currently
+      # checked out" from the pool.
+      it 'does not raise on the second call' do
+        server = authorized_client.cluster.next_primary
+        server.with_connection do |connection|
+          session.pin_to_server(server)
+          session.unpin(connection)
+
+          expect do
+            session.unpin(connection)
+          end.not_to raise_error
+        end
+      end
+    end
+  end
 end