Add softplus and softsign

[Shashi456]

We need these operators for other loss functions like kullback-leibler divergence and logcosh.

One small doubt, The comments for softplus and softsign both mention gradients as
gradients: The backpropagated gradients to the corresponding softplus operation.

So am i getting the logic right by passing gradients: v?

Tests pass.

[Shashi456]

@saeta sorry, I was probably copying them for reference but didn't remove them.

[eaplatanios]

@Shashi456 @saeta softplus is also added in #148.

[Shashi456]

@rxwei should I remove softplus from this ? Also #148 takes the derivative as sigmoid, I know it's right. But is there a good reason to not use softplusGrad given within the bindings?

[Shashi456]

@rxwei could you put a build tag on this or are you waiting for #148? In that case I'll remove softplus from here and let this PR go ahead with adding softsign.

[Shashi456]

Sorry, made some changes for build errors but did not push them till now. The build and test should pass now

[Shashi456]

@rxwei :P, Can you trigger a build once more?

[Shashi456]

@rxwei so the build passes, the error is from move, was the toolchain updated for that?

[Shashi456]

Requesting a build on this.

[rxwei]

Thanks!