handle multiple operator sockets more correctly

make node rebuilding function more reusable

Merge branch 'functions' into simulation-tree

Merge branch 'master' into functions

write node storage to file and read back

Have to find a better way to do this. An issue is that `node->typeinfo`
when the node is just read.

initial operator socket callback

run node declaration function only once per update

simplify SocketDecl

new operator socket

first node with a variable number of sockets

keep socket values after node rebuild

implicitely convert rgba_f to const float *

remove invalid links and maintain links after rebuild

access node identifier

automatically remove links between incompatible links

store incident links in virtual node tree

Merge branch 'functions' into simulation-tree

Merge branch 'master' into functions

Use new LinearAllocatedVector in various places

new LinearAllocatedVector<T>

This is a vector that does not own the memory its elements are stored in.
Instead, a linear allocator is passed to every call to `append`, that will
be used if the vector has to grow. Since a linear allocator does not support
deallocation, no memory is dealloced until the linear allocator is destructed.

This structure can be used when many small vectors of initially unknown
size are required.