[OPS] add bmm_transpose ops (#3990)

### What this PR does / why we need it?
Add a new fusion ops to custom_op, which can cobime the torch.bmm() and
transpsose to achieve better peformance. This ops is used in mla_v1 to
replace the bmm and transpose

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?


- vLLM version: v0.11.2

---------

Signed-off-by: hust17yixuan <303660421@qq.com>

csrc/kernels/math_utils.h

@@ -0,0 +1,15 @@
+#ifndef KERNEL_MATH_UTILS_H
+#define KERNEL_MATH_UTILS_H
+#include <cstdint>
+
+namespace device_utils {
+
+template <typename T, T roundVal>
+__aicore__ __force_inline__ T RoundUp(const T &val)
+{
+    return (val + roundVal - 1) / roundVal * roundVal;
+}
+
+};  // namespace device_utils
+
+#endif