Added support for micro/macro averages multi-class classification metrics

[renatopanda]

Just added support for multi-class classification metrics such as:

• Precision (micro and macro averages)
• Recall (micro and macro averages)
• F1 Score (micro and macro averages)

As described for instance here. This is directly related with the enhancement I've requested in #2

Hope everything is ok, first time editing an R package. Also there might be a bug (or is it expected?) with Confusion_DF/Precision/Recall when the positive label is missing from the predicted labels.

[renatopanda]

Added a small optimization to classification metrics. Based on this http://stackoverflow.com/questions/27303534/faster-way-to-subset-on-rows-of-a-data-frame-in-r

It still made some difference in my tests:

Unit: milliseconds
          expr      min       lq     mean   median       uq      max neval
     PRECISION 2.454476 2.913534 3.102103 3.052503 3.188008 12.70121  1000
 PRECISION_OPT 1.869344 2.271045 2.431992 2.373443 2.470644 12.88946  1000
       P_MICRO 4.902790 5.571457 5.937313 5.819753 6.042065 15.27695  1000
   P_MICRO_OPT 2.521458 2.976474 3.184339 3.109862 3.228235 12.64501  1000

[sainathadapa]

@yanyachen Hi, I'm interested in the functionality that this PR adds. Can you review this pull request and merge it? If necessary, I can help with validating the code, or adding few tests.