dlib 19.7 the padding code was changed and accidentally doubled the size of the
applied padding when in the older (and still default) landmark_relative padding
mode.  It's not a huge deal either way, but this change reverts back to the
intended behavior.

output correlation rather than squared correlation.

* disjoint_subsets: make clear and sizde functions noexcept

* disjoint_subsets: add get_number_of_sets function, documentations and tests

* disjoint_subsets: add get_size_of_set function, documentation and tests

* Split disjoint_subsets in a hierarchy.

Modify disjoint_subsets to make it a valid base class.
Add new class disjoint_subsets_sized, with information about number of
subsets, and size of each subset.
Add necessary unit test for the new class.

* Replace tabs by spaces.

* Remove virtuals from disjoint_subsets, and modify
disjoint_subsets_sized.

Now disjoint_subsets_sized is implemented in terms of disjoint_subsets,
instead of inherit from it.

functions very much when using the dnn tools, since otherwise it will sometimes
take hours to compile.

precision by default.  If the user is OK with less precision they can change
it.

and BOBYQA.  This greatly simplifies the example.

that obviously need it.

particular, rather than just dumping exactly 400 of the last loss values, it
now dumps 400 + 10% of the loss buffer.  This way, the amount of the dump is
proportional to the steps without progress threshold.  This is better because
when the user sets the steps without progress to something larger it probably
means you need to look at more loss values to determine that we should stop,
so dumping more in that case ought to be better.

* imglab: select next/previous image with 's' and 'w'

* imglab: make 'w' and 's' keys behave like UP and DOWN keys; add 'about' text