Allow input nodes inside groups to be randomised

[sfmig]

Input nodes are defined as nodes in the graph with no input sockets. The input nodes to randomise are identified because their name is prefixed with a prefix (typically 'random', case insensitive).

With this PR both independent nodes, and nodes inside a group are searched. The 'artificial' nodes that show up inside a node group, usually named 'Group input' or 'Group output' are excluded from the search.

Will close #8

[sfmig]

One potential problem with this PR as it is (probably good to discuss tomorrow with @tdowrick):

• when a node is copied, in practice we are creating a new pointer to the original node.
• If you copy a node and paste it in a new material, the values to randomise will point to the ones in the original node, but the min/max values to randomise between will be independent of the original node....
• I understand that this is actually a likely scenario right? If so, I'd suggest to fix it in a new PR

[ruaridhg]

LGTM, there may be a repeated for loop in operators that is redundant but looks sensible apart from that

[ruaridhg]

LGTM