Enable w/a also for float dtype

Author: antonwolfy

dpnp/dpnp_utils/dpnp_utils_linearalgebra.py

@@ -899,11 +899,7 @@ def dpnp_matmul(
                 # workaround with extra copying of an input array in case when
                 # it has a small size and non-zero offset
                 # TODO: remove the workaround once OneMKL issue is resolved
-                if (
-                    compute_dtype != dpnp.float32
-                    and bi._is_lnl_arl_architecture(exec_q.get_sycl_device())
-                ):
-
+                if bi._is_lnl_arl_architecture(exec_q.get_sycl_device()):
                     def _need_to_copy(a):
                         a_usm = dpnp.get_usm_ndarray(a)
                         if a_usm._element_offset > 0 and a_usm.size < 16: