This is handled on an upper layer.

This method is called from unpack which should return the object
loaded. Before the result was none.

Now both invalid case are treated as exception:
- Calling a loop using algorithm without loop_channel
- Calling a non loop using alggorithm with a loop_channel

Now the configuration includes both the "loop using" algorithm info
and the loop itself info.

The tuple should contain a:
- boolean for success or failure
- dictionary with the data to return to the algorithm

The format is:

"loop": {
    "input": {
        "type": "data_format_name"
    },
    "output": {
        "type": "data_format_name"
    }
}

The format is:

"loop": {
    "input": {
        "type": "data_format_name"
    },
    "output": {
        "type": "data_format_name"
    }
}

- loop is the validation algorithm
- loop_user is the algorithm communicating with the loop

This will call the validation method of the loop algorithm.

To distinguish from classic algorithms.

This an algorithm that must implement a validate method
that will do the analysis of the output and provide a
boolean feedback. It might also write additional information
on an output.

This is an additionnal communication channel needed for
the loop to inter-operate with the algorithm block.

This allows to have a different process method signature
for algorithm using a loop block and no need to modify
existing algorithms.

It contains custom exceptions used in the package.

Fix lazy loading of algorithm's runner

See merge request !25