It allows tools which expect version in form major.minor to work
correctly even in case when pmdk is downloaded as zip archive.

This patch also removes '%d' from GIT_VERSION file. This placeholder
was filled with name of the branch only for the HEAD of a branch.
This means that GIT_VERSION of a selceted revision had different content
(either only commit hash or commit hash + name of the branch) depending
on whether selected revision was HEAD of the branch or not. Different
content of GIT_VERSION affected hash of the archive (and caused problems
with vcpkg in the past).

Version with two dashes (like 1.5-19-gb8f78a329) confuses rpmbuild in
recent Fedoras, which makes it generate uninstallable -devel packages
(libpmem-devel doesn't have Provides: pkgconfig(libpmem) tag, which
doesn't satisfy Requires: pkgconfig(libpmem) in
libpmem[obj|blk|log]-devel).

"git -C" was added in git 1.8.5, which is not available in RH 7.x.

There are many problems:

On Linux we were using git commit id of the parent repository if
zip/tar.gz package was uncompressed inside of git repository (and
this is the case for Fedora).

We can't rely on GIT_VERSION for releases, because its contents depends
on state of the branch the tag was created from.
This file contains all refs that point to the tagged commit, so
immediately after 1.4.1 release it contained both "1.4.1" and
"stable-1.4", but after merging the first PR to "stable-1.4" branch it
contained only "1.4.1". This means that the checksum of the released
version changed! This is why vcpkg build from master fails now (it
contains initial checksum).
The fix is to remove the automatically filled GIT_VERSION file and add
static VERSION file before the release and revert this commit after.

Version reported by debug libraries was empty if libraries were built
from directories other than / and /src.

On Windows reported version for point releases was wrong.

This reverts commit db848041.

Scripts must start with #!/usr/bin/env <shell> for portability.
Add set -e to top-level scripts.
Add use warnings to perl scripts.

There are a few tests that may take more than a minute to complete,
when run on a non-pmem filesystem.  This patch extends time out
for those tests to 10 minutes.  Default timeout for other tests
is 3 minutes.

Loading symbols from librpmem was missing.

Ref: pmem/issues#499

The new way of setting a test requirements is by using a config.sh file located
in a test sub-directory. The config.sh file structure is described in
src/test/config.sh_example. It is the recommended method but it is still under
development and does not cover all possible requirements. The old way of setting
test requirements using require_* functions is still valid and can be mixed with
the new way.

Currently libpmemobj.h is a huge mess:
- it includes APIs that are incompatible (atomic vs transactional)
- it mixes core library APIs with usability wrappers
- it pollutes global namespace with macros that C++ API does not need
- there's no way for application to import only part of the API;
  eg. when application uses only transactional API it can't opt out
  of atomic API (and it can use it by accident, see commit
  5f462f7f for an example)

This commit creates new libpmemobj directory with these files:

* atomic_base.h
* atomic.h
* base.h
* iterator_base.h
* iterator.h
* lists_atomic_base.h
* lists_atomic.h
* pool_base.h
* pool.h
* thread.h
* tx_base.h
* tx.h
* types.h

libpmemobj.h includes all those files.

_base.h headers are supposed to be used only by our C++ headers.

C applications should still include libpmemobj.h - the split should
be considered an implementation detail.

Update all copyright headers:
- remove "(c)" string
- replace "Intel Corporation" with
  "the copyright holder" in the 3rd clause
- update copyright date to 2016

The additional templated operator overloads are in a separate header.
This was done to ease the integration of existing applications.

This patch introduces C++ for libpmemobj. For now
it includes 'resides on pmem' template property
class and a persistent smart pointer.

TEST0 failed because it couldn't allocate root object (chunks from
previous pool are still used after 2nd pmemobj_create).
TEST1 crashed on 2nd pmemobj_create (in heap_populate_buckets
"run" lies outside of the pool).

Not needed after previous commit.

- new type safety macros with named unions
- atomic allocation API changes:
-- remove pmemobj_alloc_construct()
-- change return values to int
-- return PMEMoid using reference
-- change PMEMoid atomically using redo log
- add TX_NEW, TX_ZNEW, POBJ_NEW and POBJ_ZNEW macros
- add size parameter to TX_ALLOC, TX_ZALLOC, POBJ_ALLOC and POBJ_ZALLOC
- update manpage

These functions use non-temporal instructions.

Ref: pmem/issues#12