Improve dialog2flow clustering handling with big datasets

[sergioburdisso]

Dialog2Flow-based components relay on building the graph, however, when the number of dialogues is larger than 2000, the clustering takes too much time, this is something that must be improved. Some ideas:

• Set a maximum number k as part of the class construction, if len(dataset) > k then we sample k dialogues.
• Use a more efficient clustering algorithm.