Implementation of Weighted CRF Tagger (handling unbalanced datasets)

[eraldoluis]

Closes allennlp issue #4619.

Depends on allennlp PR #5676

Changes proposed in this pull request:

• I implemented and experimentally compared three sample weighting strategies for CrfTagger.
• I added two parameters to CrfTagger: label_weights and weight_strategy.
• The parameter label_weights is a Dict[str, float] with a mapping {label : weight} to be used in the loss function in order to give different weights for each token depending on its label.
• The parameter weight_strategy can be: None 'emission', 'emission_transition' or 'lannoy'.
• If label_weights is given and weight_strategy is None or 'emission', then the emission score of each tag is multiplied by the corresponding weight (as given by label_weights).
• If emission_transition, both emission and transition scores of each tag are multiplied by the corresponding weight.
• If weight_strategy is 'lannoy', then we use the strategy proposed by Lannoy et al. (2019).
• An experimental comparison among these three strategies and a brief discussion of their differences here.
• Tests were created to cover the new feature.

[epwalsh]

LGTM!