fix: error for large sample sizes in Multinomial NB

I was seeing SystemStackError issues with Multinomial NB when running
with larger sample sizes. I believe this is due to splatting the
samples into `Numo::DFloat` as params. That param list can be very
large with large sample sets, which overflows Ruby's default heap
size.

I think instead we can use `Numo::DFloat.cast` here, passing in the
array, as that will just be a single param.

I've also lifted this cast to the `bin_x` variable so we don't have to
do it per class.

Author: keegnotrub

rumale-naive_bayes/lib/rumale/naive_bayes/multinomial_nb.rb

@@ -65,9 +65,9 @@ def decision_function(x)
         x = ::Rumale::Validation.check_convert_sample_array(x)
 
         n_classes = @classes.size
-        bin_x = x.gt(0)
+        bin_x = Numo::DFloat.cast(x.gt(0))
         log_likelihoods = Array.new(n_classes) do |l|
-          Math.log(@class_priors[l]) + (Numo::DFloat[*bin_x] * Numo::NMath.log(@feature_probs[l, true])).sum(axis: 1)
+          Math.log(@class_priors[l]) + (bin_x * Numo::NMath.log(@feature_probs[l, true])).sum(axis: 1)
         end
         Numo::DFloat[*log_likelihoods].transpose.dup
       end

rumale-naive_bayes/spec/rumale/naive_bayes/multinomial_nb_spec.rb

@@ -51,4 +51,16 @@
     expect(probs.shape[1]).to eq(n_classes)
     expect(predicted_by_probs).to eq(y)
   end
+
+  context 'with large sample sizes' do
+    let(:x) { Numo::DFloat.new(1_000_000, 4).rand }
+    let(:y) { Numo::Int32.new(1_000_000).rand(-1, 1) }
+
+    it 'does not raise SystemStackError', :aggregate_failures do
+      expect do
+        estimator
+        score
+      end.not_to raise_error
+    end
+  end
 end