Fixes DOOCS server shutdown lock-up

Added persistency for D_spectrum properties. This resolves #11 partially only, since there is no option to switch it off. I am not sure if that is needed though.

Issue:
Server throws when there are no device module variables:

terminate called after throwing an instance of 'std::logic_error'
  what():  ReadAnyGroup has no element with AccessMode::wait_for_new_data.
Aborted (core dumped)

This commit introduces a workaround as proposed by jens:
  void DoocsUpdater::updateLoop(){
       if (_elementsToRead.empty()) {
           return;
       }

This commit is intended to be reverted if the above strategy has side
effects.

- Replace the std::reference_wrapper<TransferElement> with a TransferElementAbstractor.

Fixed a type confusion between int64_t (aka long int on 64 bit platforms) and long long int. The types are not identical (in the C++ sense) but bit-compatible. ChimeraTK expects int64_t while DOOCS expects long long int.

fixed another uncought exception when converting e.g. an Inf from double to float (was still occurring for D_spectrum)

Do not stop server (even through assert -> undefined behaviour without Debug flags!) when trying to create a string array. Instead the property is simply ignored and a warning is printed.

Temporary roll-back to the periodic update polling instead of the readAny()-based update logic. This is necessary, since the readAny() implementation seems to be too slow for bigger servers like the LLRF server. A proper solution should be found to improve the performance of readAny() for larger number of variables.

removed the not necessary experimental switch. The readAny in control system adapter does not have the flag because it is the default behaviour

- fixed last tests which were not using timeout based checks

introduced flag to indicate when doocs initialisation is done. Removed all sleeps in the server based tests.

Removed update functions completely. The long sleep made the server unresponsive. Instead the call to run update in the tests was replaces with a sleep. Sometimes fails, as expected with a fixed timeout, but it was as hack anyway.

changed updater to run in its own thread. Tests still using doocs update, which is just doing an insanely long sleep of 1 second.

separated DoocsSpectrum into header and cc file, as it should be. It was still header only from the time where it was a tempates class.

all EqFcts are using common updater. Tests passing, but locking is not done properly across EqFcts (race conditions with RPC calls possible).

changed DoocsFactory to not create unnecessary decorators of the wrong type when auto-creating variables. Test for spectra still failing

Updater is using a vector of callbacks now. Not working as intended because there are different decorators where they should be the same.

library compiles and tests are passing, except for duplicate test which has to be adapted and is not implemented yet

changed interface of VariableMapper (list of descriptions instead of map because concept of 1:1 mapping does not work). Class compiles, but the rest of the lib incl. tests is not adapted yet

- Changes update to readLatest to allow consistent testing