Add comment describing ninv implementation trick

examples/naive/aarch64/intt_dilithium_1234_5678.s

@@ -480,6 +480,9 @@ layer1234_start:
         str_vo data14, in, (14*(512/8))
         str_vo data15, in, (15*(512/8))
 
+        // Scale half the coeffs by 1/n; for the other half, the scaling has
+        // been merged into the multiplication with the twiddle factor on the
+        // last layer.
         mul_ninv data0, data1, data2, data3, data4, data5, data6, data7, data0, data1, data2, data3, data4, data5, data6, data7
 
         canonical_reduce data0, modulus_half, neg_modulus_half, t2, t3

examples/naive/aarch64/intt_dilithium_1234_5678_manual_ld4.s

@@ -478,6 +478,9 @@ layer1234_start:
         str_vo data14, in, (14*(512/8))
         str_vo data15, in, (15*(512/8))
 
+        // Scale half the coeffs by 1/n; for the other half, the scaling has
+        // been merged into the multiplication with the twiddle factor on the
+        // last layer.
         mul_ninv data0, data1, data2, data3, data4, data5, data6, data7, data0, data1, data2, data3, data4, data5, data6, data7
 
         canonical_reduce data0, modulus_half, neg_modulus_half, t2, t3

examples/naive/aarch64/intt_dilithium_123_45678.s

@@ -514,6 +514,9 @@ layer123_start:
         str_vo data6, in, (6*(1024/8))
         str_vo data7, in, (7*(1024/8))        
 
+        // Scale half the coeffs by 1/n; for the other half, the scaling has
+        // been merged into the multiplication with the twiddle factor on the
+        // last layer.
         mul_ninv data0, data1, data2, data3, data0, data1, data2, data3
 
         canonical_reduce data0, modulus_half, neg_modulus_half, t2, t3

examples/naive/aarch64/intt_dilithium_123_45678_manual_ld4.s

@@ -523,6 +523,9 @@ layer123_start:
         str_vo data6, in, (6*(1024/8))
         str_vo data7, in, (7*(1024/8))        
 
+        // Scale half the coeffs by 1/n; for the other half, the scaling has
+        // been merged into the multiplication with the twiddle factor on the
+        // last layer.
         mul_ninv data0, data1, data2, data3, data0, data1, data2, data3
 
         canonical_reduce data0, modulus_half, neg_modulus_half, t2, t3

examples/naive/aarch64/intt_kyber_123_4567.s

@@ -437,6 +437,9 @@ layer123_start:
         str_vo data6, in, (6*(512/8))
         str_vo data7, in, (7*(512/8))
 
+        // Scale half the coeffs by 1/n; for the other half, the scaling has
+        // been merged into the multiplication with the twiddle factor on the
+        // last layer.
         mul_ninv data0, data1, data2, data3, data0, data1, data2, data3
 
         str_vi data0, in, (16)

examples/naive/aarch64/intt_kyber_123_4567_manual_ld4.s

@@ -432,6 +432,9 @@ layer123_start:
         str_vo data6, in, (6*(512/8))
         str_vo data7, in, (7*(512/8))
 
+        // Scale half the coeffs by 1/n; for the other half, the scaling has
+        // been merged into the multiplication with the twiddle factor on the
+        // last layer.
         mul_ninv data0, data1, data2, data3, data0, data1, data2, data3
 
         str_vi data0, in, (16)