numeric.c: optimize `float ** 2` case by fastpath

It would be a relatively frequent case.  It is still slower than
`float * float` because `*` has a dedicated VM instruction (opt_mult),
though.

numeric.c

@@ -1309,7 +1309,11 @@ VALUE
 rb_float_pow(VALUE x, VALUE y)
 {
     double dx, dy;
-    if (RB_TYPE_P(y, T_FIXNUM)) {
+    if (y == INT2FIX(2)) {
+	dx = RFLOAT_VALUE(x);
+        return DBL2NUM(dx * dx);
+    }
+    else if (RB_TYPE_P(y, T_FIXNUM)) {
 	dx = RFLOAT_VALUE(x);
 	dy = (double)FIX2LONG(y);
     }

test/ruby/test_float.rb

@@ -305,6 +305,7 @@ class TestFloat < Test::Unit::TestCase
     assert_equal(1.0, 1.0 ** (2**32))
     assert_equal(1.0, 1.0 ** 1.0)
     assert_raise(TypeError) { 1.0 ** nil }
+    assert_equal(9.0, 3.0 ** 2)
   end
 
   def test_eql