detail::pair: make piecewise ctor not constexpr

It should not be marked as constexpr because it does not
initialize all members directly (but through another ctor).

std::pair piecewise ctor is constexpr only in c++20.

Disable false-postivie helgrind tests

Disable the concurrent_hash_map_rehash_mock tests
due to intermittent failures (most probably false-positive
reported by helgrind).

Ref: 53c8568c
Ref: #469

Use https:// where appropriate.

concurrent_hash_map fixes

Merge stable 1.8 into master

* std::pair<> does not guarantee stable layout
* we need to have constructor which takes only one value and
  default intializes the second value in pair (it's needed for
  insert which only takes key)

Add more defrag tests

The old version lived in PMDK's space.

Add defrag to concurrent hash map

After introducing defrag() compiltaion of test with 1.8 headers fails.

Set tests' timeout value to 590 secs,
because after 10 minutes the build is considered as stalled by Travis:

"No output has been received in the last 10m0s,
this potentially indicates a stalled build
or something wrong with the build itself."

It's needed to iterate over all internal pointers of basic_string.
It's used in the 'defrag' class for storing all pointers
in the internal list for the defragmentation process.

It is more obvious that on_init_size is always consistent this way.

to runtime_initialize()

So that the initialization is only in one place.

in runtime_initialize

It will protect user in case of accidentaly changing size of key
or value.

tests: fix transaction test

detect GDB and disable tests

We did not set counter to a proper value which caused failure on
older std libs (for example in rhel-7.7)

ref issue #469

Add simple hash_map benchmark for opening time

concurrent_hash_map: change implementation of runtime_initialize(bool)

This patch changes implementation to be the same as before
708bd019 (implementation of
runtime_initialize(void) is left unchanged).

The change is needed for backward compatibility with pmemkv-1.0. There, we
called runtime_initialize(true) in a transaction after tx commit (we assumed
there is nothing transactional in runtime_initialize). Introducing a tx
inside new runtime_initialize() results in crash on 1.0