WT_DUPLICATE_KEY does not cause other threads to roll back

[JackLiJXL]

this may result in inconsistent behaviour in the following case:
read_optimized = false
node a, node b already exist in database

add_edge(a, b)
delete_node(a)

called concurrently

add_edge(a, b) will not cause delete_node(a) to roll back in this case, a possible outcome when threads finish executing is (a, b) is added but a no longer exists

[JackLiJXL]

fix: get and put back the same value with update on duplicate key error

[JackLiJXL]

need further testing to determine if this is a real problem