* the user can specify one of {none,AVX,AVX2,KNL} as
  targets for vectorization
* if modcc is available in PATH it will be used, which
  makes cross compilation easier
* a new flag USE_OPTIMIZED_KERNELS can be set to
  generate optimized kernels from the modcc compiler
  (equivalent to adding -O flag to modcc)

* `nest::mc::threading::sort` is analogous to `std::sort`
    * when using TBB the `tbb::parallel_sort` is used
    * otherwise `std::sort` is used
* used to parallelise connection construction during setup
    * now all initialization and simulation is multi-threaded
* added a unit test

* replace a call to `std::iota` with a for loop because the Intel
compiler was optimizing out the `std::iota` call in a unit test.
* add uniform syntax for converting pointer+size and ArrayView+size
into an ArrayView to provide uniform method for accessing ranges
inside mechanisms with/without optimizations.

* use iterator and const_iterator type members instead
of iterator_type and const_iterator_type
* add begin() end() to the thread_enumerable spike storage
to iterator over thread local storage

Command-line support for limitting trace output.

New options:
 * -p, --probe-ratio      Attach probes to this proportion of cells
 * -X, --probe-soma-only  Don't probe dendrites
 * -P, --trace-prefix     Specify where to write traces
                          (default: 'trace_')
 * -T, --trace-max-gid    Don't trace cells with a larger gid than this.

Fixes issue #40

The resulting trees are not equivalent to the initial ones. All
offending branches were split and their non-contiguous part was moved
directly below the root.

asynchronous communication and compute 

Feature/model refactor

* fixed const interface to the serial proxy for thread private storage
* added a serial proxy for tbb task group and moved the tbb task group
  behind the threading interface.
* added comments to the communicator to better document each method

- moved spike source storage to its own file
- added unit tests for spike storage

Merging for PR#64 review.

Address PR#64 review comments.

- communicator no longer accepts or stores thread-local spike buffers
- buffers and queues are stored and managed outside communicator
- a new type thread_private_spike_store that manages the thread
  private spike buffers

Feature/model+recipe

Regression test for bugfix in commit #712c00a7

Incorrect source ids were being given to detectors;
targets were then rightly ignoring the generated spikes.

Also: small coding style compliance edits.

fix compiler warnings and tests with ASSERTIONS on

- update modparser external repo to a version that fixes some compiler
  warnings with clang
- fix a handful of compiler warnings for signed-unsigned comparison in
  tests and when ASSERTIONS are on
- comment out EXPECTS(has_contiguous_segments) where they fail in the
  tests so that we can run all the tests with full debug flags

Bugfix: build fail without TBB.