Changed the ROC curve to implement what is required by 'The Handbook of Face Recognition'

Closes #24 

I have implemented the correct way to compute FRR values based on The Handbook of Face Recognition, i.e., counting positive scores that are exactly at the threshold to be correctly classified.

See merge request !24

Merge branch '23-typeerror-underlying-read-should-have-returned-a-bytes-like-object-not-str' into 'master'

Fixed issue of TextIO around TextIO

Closes #23 

Without understanding, where the error comes from, I provide a work-around that should work.

See merge request !23

Make use of test set on compute_perf optional; Update docs

This MR addresses two issues:

* Users which don't have a test set can now also use `bob_compute_perf`. In this case, only reduced number of statistics and plots are printed
* Fixes to the user guide, mainly concerning how to call the scripts since the simplifications with docopt

See merge request !22

Fixes verbosity; Fixes CMC plot (closes #20)

See merge request !21

Resolve "load_scores extremely memory hungry"

Closes #19 

Currently I left the old functions to read scores inside, but it might be a good idea to remove them and base everything on iterators. What do you think?

See merge request !20

[Automatic] update links and the ci mostly



See merge request !18

[Automatic] update links and the ci mostly



See merge request !17

Add score distributions on compute_perf (implementation similar to @akomaty's) - closes #15



See merge request !16

Docfix

This fixes include migrating from argparse to docopt and huge simplifications on the command-lines for each application.

It closes #16.

See merge request !15