[Example] Add the fused kernel for decoding of flash linear attention

[yaoyaoding]

This PR adds the example of fused kernel for flash linear attention:

Sigmoid Gating Delta Rule Update Benchmark Results:
      name     (B, T, H, K)   (HV, V) latency (ms)
0    torch   (1, 1, 4, 128)  (8, 128)        1.164
1   triton   (1, 1, 4, 128)  (8, 128)        0.010
2    tilus   (1, 1, 4, 128)  (8, 128)        0.006
3    torch   (1, 2, 4, 128)  (8, 128)        2.270
4   triton   (1, 2, 4, 128)  (8, 128)        0.011
5    tilus   (1, 2, 4, 128)  (8, 128)        0.007
6    torch   (1, 4, 4, 128)  (8, 128)        4.475
7   triton   (1, 4, 4, 128)  (8, 128)        0.013
8    tilus   (1, 4, 4, 128)  (8, 128)        0.009
9    torch   (1, 8, 4, 128)  (8, 128)        8.848
10  triton   (1, 8, 4, 128)  (8, 128)        0.018
11   tilus   (1, 8, 4, 128)  (8, 128)        0.013
12   torch  (1, 16, 4, 128)  (8, 128)       17.589
13  triton  (1, 16, 4, 128)  (8, 128)        0.029
14   tilus  (1, 16, 4, 128)  (8, 128)        0.022
15   torch  (1, 32, 4, 128)  (8, 128)       35.413
16  triton  (1, 32, 4, 128)  (8, 128)        0.051
17   tilus  (1, 32, 4, 128)  (8, 128)        0.044