[FIX][PY] Repair tensorize PrimExpr dtype access

Author: tqchen

tests/python/s_tir/schedule/test_tir_schedule_tensorize.py

@@ -842,7 +842,7 @@ def _tir_packed_int_to_int_to_float(storage_nbit: int):
     storage_dtype = "int" + str(storage_nbit)
 
     def f_convert(nbit: int, val: tirx.PrimExpr, pos: tirx.PrimExpr, dtype: str):
-        assert val.dtype == storage_dtype
+        assert val.ty.dtype == storage_dtype
         mask = tirx.const((1 << nbit) - 1, "int32")
         unextended = (val >> (pos.astype("int32") * tirx.const(nbit, "int32"))) & mask
         return tirx.Cast(dtype, (unextended << tirx.const(32 - nbit, "int32")) >> tirx.const(32 - nbit, "int32"))