Adds support for softmax

[mxmasterton]

Implements exp and log operations as well as softmax

[54g0]

It is wrong the backward function inside the log function is wrong because the derivative of logx is 1/x so self.grad += (1/self.data)*out.grad

def log(self):
    out = Value(math.log(self.data),(self,),"log")
    def _backward():
        self.grad += (1/self.data)* out.grad
    out._backward = _backward
    return out 

and one more bug you by default python start counting from the integer 0 so it tries to do 0 + Value() so Value class doest know how to add a plain integer from the left side so this would fail the correct way is denom = sum(exps,Value(0.0))

@staticmethod
    def softmax(logits):
        exps = [logit.exp() for logit in logits]
        denom = sum(exps,Value(0.0))
        return [e/denom for e in exps]

[mxmasterton]

Will fix the first bug in a second. Second one is not an issue because radd is implemented and also there is a check right at the start of add to catch this.