#0: Add the dim 0 support repeat backward

[ruthreshx]

Added the dim 0 support for repeat backward.
yet to add the support for dim 1.

[ruthreshx]

Hi @tarafdarTT ,

• With the help of a link I have derived the repeat backward formula in python & it is working as we expected in colab for the various inputs.

• The bottleneck is, the intermediate grad_size and sum_dims are incompatible for the TT support. It requires >4D dims to reshape and sum accordingly.

• Due to this we were unable to add the entire support, but we were able to add the partial support for the dim 0 & 1, only when N and C is 1 and the repeat is of any number for such dim, other should be 1.

• For the dim 2 & 3, it requires the shape should TILE, rather our one of the constraints is, the respective dim should be 1.

Supported shapes
Ex:

Input shape:
 [1, 1, 32, 32], [1,1, 320, 320] (N & C should be one)
Repeats: 
 [12, 1, 1, 1], [6, 1, 1, 1], [1, 24, 1, 1], [1, 3, 1, 1]]  (repeat is of any number for such dim, other should be 1).

          For dim 0, Input is [1,1, 32, 32] the repeats for such dim should be any number.
	  [32, 1, 1, 1], [2, 1, 1, 1], [6, 1, 1, 1]
           
	 For dim 1, Input is [1,1, 64, 64] the repeats for such dim should be any number.
	  [1, 24, 1, 1], [1, 30, 1, 1], [1, 14, 1, 1]

For more info, please find the below doc link
https://docs.google.com/document/d/1kjQPPqD8uG-M14qz5AJlWoSY_otV6u6TJGvw9LvQPsE/edit

[ntarafdar]

@ruthreshx this is awesome ty! This LGTM, I see you have added support at the tt_lib level. In a future PR we could look at adding this at the ttnn level as well.

[tt-aho]

All the TT_ASSERT should be switched to TT_FATAL here

[ruthreshx]

Fast dispatch: https://github.com/tenstorrent-metal/tt-metal/actions/runs/8453688540
Slow dispatch: https://github.com/tenstorrent-metal/tt-metal/actions/runs/8453708409