Add additional fields in batch reader, run_model output file & knowledge distillation data files

[haowu666]

[The updated version] generalized the previous version (adding post_id). The main workflow can read list of addtional fields such as post_id, page_id, page_url into context, then parse it along with classification metric reporter. Now PyText users are able to include these fields in the file of saving results, and it's useful to look into the details of the model output, explore other metric assocaited with team target, or look for other relevant info form hive table by searching on the fileds. Also, in knowledge distillation, each gen_kd_data[i] where i in {0,1,2} (training, validation, test) can handle label list for multi-label task, and include the additional fields as a dictionary in the generated data files, which is helpful for building teacher and student network for multi-label experiments.

In the debug file for the test dataset, one header example is
#predicted, actual, scores_str, text, post_id, post_url

In generating knowledge distillation data for training dataset, one header example is
#label_list, score, logit, label_names(in order), text, {"post_id": 123456, "post_url": http://post_url.com}

[The previous version] aimed at reading post_id and parsing it along with the output file, so that the file is useful for our metric calculation and building KD - teacher/student in our workflow in the following steps. Other users that have the same desired goal can use this Diff as a potentially alternative solution.

Differential Revision: D16271134

[facebook-github-bot]

This pull request has been merged in e14ea15.