BiDirectional: Custom REFs resulting consumed_resources is wrong

[torressa]

Describe the bug When running the BiDirectional algorithm in both directions and using custom REFs, the resulting consumed_resources attribute is wrong.

[torressa]

This is due to the label inversion being a bit trickier with custom REFs. I've implemented a really inefficient solution https://github.com/torressa/cspy/blob/master/cspy/algorithms/bidirectional.py#L561-L580

The proper way to do this would be to flip the initialisation of backward labels, such that instead of decreasing from the upper bound, they increase from the lower bound. Except the monotone resource which is required to decrease. This removes the need for an inverted backward REF, which is quite convenient for everyone.

[torressa]

It turns out it's quite hard to come up with a generic way to merge forward and a backward labels when custom REFs are given. It's really hard to guarantee the correct resource consumption. It's pretty easy per problem, I just can't see a generic fix. The best solution I've come up for this is to create another input for BiDirectional that takes a function that does this.

https://github.com/Kuifje02/vrpy/issues/10#issuecomment-619194443

[torressa]

The addition of a new argument REF_join in BiDirectional removes the issue. Allowing the user to define an appropriate function for merging two https://github.com/torressa/cspy/blob/6ad0650e5baa27882594a0e4fe6a0c5835fb0504/cspy/algorithms/bidirectional.py#L82 Only required when custom REFs are used, and the joining of backward and forward paths requires some specific operations. For example, node-related resource extensions. For an example for the case of the vehicle routing with time windows https://github.com/Kuifje02/vrpy/blob/dev/vrpy/subproblem_cspy.py#L246 Docs need updating before next release