Adding backward kernel for repkv on llama3 branch (cudamode-irl)

[insop]

PTAL, repkv_backward is updated and tested.
I will update repkv.cuh once this PR is merged.

CC: @karpathy

This is an WIP repkv backward kernel, started as a cudamode-irl project.
Once the following work is done, will remove draft sign.

This work was supported by ALEKSA (@gordicaleksa) , Eric (@ngc92), ARUN (@ademeure) during the irl event.

• pytorch backward test code

• CPU kernel

• Cuda kernel

• build

make repkv_backward
/usr/local/cuda/bin/nvcc -O3 --use_fast_math --generate-code arch=compute_80,code=[compute_80,sm_80] -lcublas -lcublasLt -std=c++17 repkv_backward.cu -o repkv_backward

• test run on A30

Using kernel 1
Checking block size 32.
0.531524 0.531524
0.600285 0.600285
0.458787 0.458787
0.296680 0.296680
-0.911627 -0.911627
Checking block size 64.
0.531524 0.531524
0.600285 0.600285
0.458787 0.458787
0.296680 0.296680
-0.911627 -0.911627
Checking block size 128.
0.531524 0.531524
0.600285 0.600285
0.458787 0.458787
0.296680 0.296680
-0.911627 -0.911627
Checking block size 256.
0.531524 0.531524
0.600285 0.600285
0.458787 0.458787
0.296680 0.296680
-0.911627 -0.911627
Checking block size 512.
0.531524 0.531524
0.600285 0.600285
0.458787 0.458787
0.296680 0.296680
-0.911627 -0.911627
Checking block size 1024.
0.531524 0.531524
0.600285 0.600285
0.458787 0.458787
0.296680 0.296680
-0.911627 -0.911627
All results match. Starting benchmarks.

block_size   32 time 3.2461 ms
block_size   64 time 1.7509 ms
block_size  128 time 1.7374 ms
block_size  256 time 1.7441 ms
block_size  512 time 1.8092 ms
block_size 1024 time 2.0443 ms

[insop]

@gordicaleksa , @ngc92, @ademeure, @karpathy

PTAL, repkv_backward cpu and cuda kernels are updated and tested.
I will update repkv.cuh once this PR is merged.