Adding new ctor for `Session`

Meant to be a shortcut to directly set some basic configuration.

E.g.

```
var sessionOptions = new SessionOptions();
sessionOptions.SetConfig(new ConfigProto
{
   LogDevicePlacement = true,
   InterOpParallelismThreads = 4,
   GpuOptions = new GPUOptions
   {
      AllowGrowth = true,
      PerProcessGpuMemoryFraction = 0.12
   }
});
var session = tf.Session(sessionOptions);
```

Adding `AbsGrad`

To be aligned with Tensorflow, introducing also `Sign` operation
and its gradient.

src/TensorFlowNET.Core/APIs/tf.math.cs

@@ -203,6 +203,9 @@ namespace Tensorflow
         public static Tensor sqrt(Tensor a, string name = null) 
             => gen_math_ops.sqrt(a, name);
 
+        public static Tensor sign(Tensor a, string name = null)
+            => gen_math_ops.sign(a, name);
+
         public static Tensor subtract<T>(Tensor x, T[] y, string name = null) where T : struct
             => gen_math_ops.sub(x, ops.convert_to_tensor(y, dtype: x.dtype.as_base_dtype(), name: "y"), name);
 

src/TensorFlowNET.Core/Gradients/math_grad.cs

@@ -16,6 +16,7 @@
 
 using System;
 using System.Linq;
+using Tensorflow.Operations;
 using static Tensorflow.Python;
 
 namespace Tensorflow.Gradients
@@ -26,6 +27,15 @@ namespace Tensorflow.Gradients
     [RegisterGradient("math_grad")]
     public class math_grad
     {
+        [RegisterGradient("Abs")]
+        public static Tensor[] _AbsGrad(Operation op, Tensor[] grads)
+        {
+            var x = op.inputs[0];
+            var grad = grads[0];
+
+            return new Tensor[] { gen_ops.mul(grad, gen_math_ops.sign(x)) };
+        }
+
         [RegisterGradient("Add")]
         public static Tensor[] _AddGrad(Operation op, Tensor[] grads)
         {
@@ -428,6 +438,15 @@ namespace Tensorflow.Gradients
             });
         }
 
+        [RegisterGradient("Sign")]
+        public static Tensor[] _SignGrad(Operation op, Tensor[] grads)
+        {
+            var x = op.inputs[0];
+            var zero = constant_op.constant(0.0f, x.dtype, x.shape);
+
+            return new Tensor[] {zero};
+        }
+
         [RegisterGradient("Square")]
         public static Tensor[] _SquareGrad(Operation op, Tensor[] grads)
         {

src/TensorFlowNET.Core/Operations/gen_math_ops.cs

@@ -210,6 +210,13 @@ namespace Tensorflow
             return op.outputs[0];
         }
 
+        public static Tensor sign(Tensor x, string name = "Sign")
+        {
+            var op = _op_def_lib._apply_op_helper("Sign", name: name, args: new {x});
+
+            return op.outputs[0];
+        }
+
         public static Tensor sinh(Tensor x, string name = null)
         {
             var _op = _op_def_lib._apply_op_helper("Sinh", name, args: new { x });

src/TensorFlowNET.Core/Operations/math_ops.cs

@@ -216,6 +216,15 @@ namespace Tensorflow
             return gen_math_ops.sigmoid(x_tensor, name: name);
         }
 
+        public static Tensor sign(Tensor x, string name = null)
+        {
+            return with(ops.name_scope(name, "Sign", new {x}), scope =>
+            {
+                x = ops.convert_to_tensor(x, name: "x");
+                return gen_math_ops.sign(x);
+            });
+        }
+
         /// <summary>
         /// Returns (x - y)(x - y) element-wise.
         /// </summary>

src/TensorFlowNET.Core/tf.cs

@@ -72,5 +72,10 @@ namespace Tensorflow
         {
             return new Session(graph);
         }
+
+        public static Session Session(SessionOptions opts)
+        {
+            return new Session(null, opts);
+        }
     }
 }