Prioritize gemv/gerc over dot22scalar

The marked xfail test was failing because Ger wasn't introduced, not because of the complex dtype.

Author: ricardoV94

pytensor/tensor/rewriting/blas.py

@@ -758,7 +758,7 @@ def local_dot22_to_ger_or_gemv(fgraph, node):
         ignore_newtrees=False,
     ),
     "fast_run",
-    position=15,
+    position=11,
 )
 
 
@@ -903,7 +903,7 @@ def local_dot22_to_dot22scalar(fgraph, node):
     "local_dot22_to_dot22scalar",
     in2out(local_dot22_to_dot22scalar),
     "fast_run",
-    position=11,
+    position=12,
 )
 
 

tests/tensor/test_blas.py

@@ -1903,17 +1903,9 @@ def test_f32_1_2(self):
     def test_f64_4_5(self):
         return self.given_dtype("float64", 4, 5, destructive=False)
 
-    @pytest.mark.xfail(
-        condition=config.floatX == "float32",
-        reason="GER from complex64 is not introduced in float32 mode",
-    )
     def test_c64_7_1(self):
         return self.given_dtype("complex64", 7, 1)
 
-    @pytest.mark.xfail(
-        raises=AssertionError,
-        reason="Unclear how this test was supposed to work with complex128",
-    )
     def test_c128_1_9(self):
         return self.given_dtype("complex128", 1, 9)