Replace memset with fill to work for non-trivial scalars.

For custom scalars, zero is not necessarily represented by
a zeroed-out memory block (e.g. gnu MPFR). We therefore
cannot rely on `memset` if we want to fill a matrix or tensor
with zeroes. Instead, we should rely on `fill`, which for trivial
types does end up getting converted to a `memset` under-the-hood
(at least with gcc/clang).

Requires adding a `fill(begin, end, v)` to `TensorDevice`.

Replaced all potentially bad instances of memset with fill.

Fixes #2245.

unsupported/Eigen/CXX11/src/Tensor/TensorContractionGpu.h

@@ -1370,8 +1370,8 @@ struct TensorEvaluator<const TensorContractionOp<Indices, LeftArgType, RightArgT
     // columns in right side
     const Index n = this->m_j_size;
 
-    // zero out the result buffer (which must be of size at least m * n * sizeof(Scalar)
-    this->m_device.memset(buffer, 0, m * n * sizeof(Scalar));
+    // zero out the result buffer (which must be of size at least m * n * sizeof(Scalar))
+    this->m_device.fill(buffer, buffer + m * n, Scalar(0));
 
     typedef internal::TensorContractionInputMapper<LhsScalar, Index, internal::Lhs,
                                                    LeftEvaluator, left_nocontract_t,