Made changes to allow to work with large datasets

[LeonidBeynenson]

Earlier sometimes error happened when a temporary buffer in ml module were created as a Mat,
since the length of the row of the buffer (in bytes) was sizeof(int) * (work_var_count + 1)*sample_count
-- in some cases it was greater than INT_MAX.

[vpisarev]

ok, just documenting what we discussed with Leonid.
The patch includes quite many changes and most of the changes replace the same simple formula with the same more complex formula. It's not guaranteed that:

1. all the places are modified.
2. the complex formula will work fine in future (and in this case we will need to make it even more complex in all those places?!)

So, instead I suggest to replace "CvMat* buf" with "<some_object_t> but", which will have the proper methods. It should solve both problems:

1. If we forget to replace one of "buf->cols" etc., we get compile error.
2. the complex formula will be hidden in the method and therefore we don't have to duplicate it in many places.
   We can do such a change because the patch is for master, not 2.4, and in 2.5 we can break binary compatibility.

[kirill-korniakov]

Leonid, any plans with this task? Could you please share your thoughts?

[LeonidBeynenson]

I can start work on this task, but I thought that now Vadim Pisarevsky is going to work with it.

[alekcac]

Vadim, any changes?

[kirill-korniakov]

I suppose that Leonid will finish this task. Leonid, am I right?

[LeonidBeynenson]

I will finish it when I have a time slot for it.

[mdim]

@kirill-kornyakov Kirill, when you give Leonid a time slot for this issue eventually? :)
@LeonidBeynenson Leonid, please back to this old task!

[vpisarev]

closing this one, because it's superseded by #395