No longer in use as per switch to MPI-based parallelization.

Discard phase points in *-segregate-phases if absent from SVM model

Closes #3

See merge request J.Greitemann/svm-order-params!5

This fixes #3.

If no samples have been taken at a phase point, the SVM model will not
know about it can thus no graph edges will connect to that point.
Previously, those points would still be included in the graph analysis,
resulting potentially in many disconnected points and a large degeneracy
of eigenvalue zero. The new behavior is to discard those phase points
flat out.

Restore parameters points from all merged clones in the graph analysis

Closes #2

See merge request J.Greitemann/svm-order-params!4

Rather than locating the clone files based on the `merge` parameter and
constructing the sweep policies of the merged clones directly from their
parameters, we now use the `merged_parameters` which are stored within
the *.out.h5. This way, the graph analysis is yet again dependent on
only the *.out.h5, meaning that it may be performed locally from the
downloaded out file, while the much larges clones remain on the cluster.

This addresses #2. For now, this reads the parameters from the clones
referenced in the `merged` parameter.

Adding classifier

See merge request J.Greitemann/svm-order-params!3

...in (de)serialization of procrastination_adapter (LAZY mode).
Addresses a concern raised in !3.

Previously used Simulation::configuration() led to a bug when loading from checkpoint,
as the current phase point might lead to an empty configuration. By using
random_configuration instead, it is ensured that the prototype is always a full valid
configuration.

Fixes #1.

This was a remnant from the old OpenMP-based parallelization.

Surpresses some -Wnarrowing warnings.

In some versions of ALPSCore, count() on result_type returns an int, rather
than a size_t. This change then confuses the template parameter resolution
of std::max in the reduction.

The member types `phase_label` and `phase_sweep_policy_type` formerly
required a definition as part of the TKSVM simulation class interface.
The need for this historically originated in when classifiers and sweep
policies were determined at compile time. Now that these are derived of
abstract policies, the remaining uses of these types could be replaced
by their canonical definitions:

    using phase_label = typename phase_space::classifier::policy<phase_point>::label_type;
    using phase_sweep_policy_type = phase_space::sweep::policy<phase_point>;

TKSVM-compliant simulation classes are thus now longer required to
implement these.

This enables simulation classes to intentionally return an "empty"
configuration in cases where no samples are desired but the simulation
is merely run as part of a larger PT set.

"Emptiness" is established by calling empty() on the configuration.
Since not all configuration type may implement empty(), the comparison
`size() == 0` is a compile-time fallback. If no `size()` is provided,
the distance between begin and end iterators is used.