fix image resizing dimensions for IO

'caffe test' computes and prints all scores and their names

fix memory_used_ by computing after SetUp

compute the eltwise product gradient using a slower but stabler formula.

Make eltwise product gradient stabler (fixes issues with WITHIN_CHANNEL LRN in cifar_full example)

Drop the legacy CURAND initialization steps; these are unnecessary and
cause dramatic slowdowns for CUDA 6.5. This does no harm for K20 usage
counter to the note at least for CUDA 6.5 and 5.0.

Drop obsolete CURAND reset for CUDA 6.5 compatibility

fix layer_factory.cpp bug: there should be no ifdefs

Use gflags to parse command line arguments for convert_imageset

- All the instructions mentioned for CentOS/RHEL are also valid for Fedora Linux so add it.
- The package name for atlas is atlas-devel not libatlas-devel for CentOS/RHEL/Fedora so fix it.
- Add 'sudo' prefix to missing places to be consistent with the rest of the document.
- Add instructions for installing python headers in Fedora into the python support section.
- Update Linux installation section to reflect Fedora, refactor a bit.

Back-merge these doc fixes:

  clarify project origin
  [docs] fix citation bibtex

All other changes are spurious commits from cherry-picking.

Replace pthread with boost::thread

- ignore models and snapshots, means, and input databases
- excluding exceptions to commits is easier than including exhaustively

Standardize examples

- further detail merits of fine-tuning: less starving for itme and data
- set random seed for reproducing the tutorial
- 1557 train / 382 test split is more indicative of training quality
  than splits of 200 images

Conv layer: fix crash by setting weight pointer

Fixed CMake building test objects multiple times

Fixed CMake script of FindOpenBLAS.

cuDNN acceleration

change uses of tmpnam to mkstemp/mkdtemp

linecount counts more dirs than just src/

compile on OSX w/ CUDA 6.5

This reverts commit 8819f59.

rebase and fixup BVLC#688 from @to3i: elementwise max