feature: add implementation for 'same' mode for convolve

[mrityunjai01]

1. Added ’same’ mode in scipy.signal.convolve, (in orther words, a forward pass or a direct application of convolve)
2. Added the vjp autograd function (backward or reverse differentiation)
3. Added a test for ‘same’ mode forward pass.

This should make it straightforward to write an implementation for np.convolve covering all modes.
I claim that the VJP for convolve doesn't necessarily need to be composed of other defined primitives to make it infinitely differentiable. The second derivative of convolve is zero, with respect to either argument (Claim).

Remaining tasks:

1. Add test to check for grad of convolve
2. Add jvp for convolve

The changes are written to not touch the existing code for ‘valid’ and ‘full’ modes, while minimizing the ‘git diff’ in this commit.

[agriyakhetarpal]

Thanks, @mrityunjai01! Were you able to get to writing the JVP for convolve and the tests? :)

[mrityunjai01]

thanks, wondering if it makes more sense to include jvp in a separate PR

[agriyakhetarpal]

I think that should be fine!

[mrityunjai01]

I have added the tests, they pass for both gradients and function evaluation. The PR achieves its objective features.