Update 02-fused-softmax.md

[GlowLED]

compiled kernel启动时，现在的官方文档已经是需要传入BLOCK SIZE和num stages的写法了。现在compiled kernel启动时需要传入constexpr参数。 按照原写法，启动kernel时会报错：Traceback (most recent call last):
File "/root/triton-learn/src/fused_softmax.py", line 92, in
y_triton = fused_softmax(x)
File "/root/triton-learn/src/fused_softmax.py", line 78, in fused_softmax
kernel[(num_programs, 1, 1)](
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
x,
^^
...<4 lines>...
n_cols,
^^^^^^^
)
^
File "/root/triton-learn/.venv/lib/python3.13/site-packages/triton/compiler/compiler.py", line 498, in runner
self.run(grid[0], grid[1], grid[2], stream, self.function, self.packed_metadata, launch_metadata,
~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
knobs.runtime.launch_enter_hook, knobs.runtime.launch_exit_hook, *args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/triton-learn/.venv/lib/python3.13/site-packages/triton/backends/nvidia/driver.py", line 713, in call
self.launch(gridX, gridY, gridZ, stream, function, self.launch_cooperative_grid, self.launch_pdl,
~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
global_scratch, profile_scratch, *args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: function takes exactly 21 arguments (19 given)

少了两个参数，正好是那两个constexpr。
详情见英文官方文档：https://triton-lang.org/main/getting-started/tutorials/02-fused-softmax.html