New approach is agnostic w.r.t. client codes. The `sim_base` type needs
to be defined in the main client-code include.

Sized sweep policies allow for the definition of a single
`fixed_from_sweep` classifier that works with points generated from
arbitrary sweep policies.

This is in preparation of making phase classifiers selectable at
runtime.

Classifiers now provide additional functions `name` and `size` to expose
human-readable labels.

As a consequence, the `phase_diagram` classifier could be rewritten such
that a single class covers both D2h and D3h cases, and indeed all phase
diagrams defined on the same (J1J3) parameter space.

The ISING preprocessor variable definition used to select the ising
model in the frustmag client code clashed with that which selects the
ising client code. The former has been renamed to ISING_HAMILTONIAN
(along with HEISENBERG_HAMILTONIAN) to avoid this. This change does not
affect the user-facing values of the HAMILTONIAN CMake variable.

phase_space_policy.cpp is required in case the phase_diagram classifier
is used. Omission led to linker errors.

The training_adapter no longer manages the sweep. The interface function
update_phase_point has been changed such that it accepts concrete phase
space points, rather than generating them from the sweep policy. The
sweep_policy is only used in the sample program and consumed to generate
all the phase space points.

Resuming simulation from the checkpoint in the sample program works now.

The learn program has been stripped down to its SVM optimization part;
no sampling takes place there.

cmake -DCONFIG_MAPPING=LAZY for lazy evaluation of the configuration
policy (using the new procrastination_adapter);
cmake -DCONFIG_MAPPING=EAGER (default) for the old direct evaluation of
the configuration policy upon sampling.

Added frustmag_config_policy, derived from monomial_config_policy.

New executables for svm-order-params:
* frustmag-learn
* frustmag-test
* frustmag-coeffs

Configure project for use of specific Hamiltonian and lattice, e.g.:

    cmake -DHAMILTONIAN=heisenberg -DLATTICE=kagome ..

Use a config policy (derived from monomial_config_policy) for the ising code as
well.

Retired the old ising-coeffs program in favor of one derived from the generic
coeffs program. This was done for consistency. We loose the ability to calculate
the coefficient matrix as a function of distance (exploiting translational symm;
see PonteMelko17). Higher-rank monomial kernels can be used (which is pointless
of course for the Ising model, but still).

Moved stuff not related to the gauge model to the root directory.

config_policy.hpp holds the abstract config_policy interface, the (reusable)
monomial_config_policy and the dummy block_config_policy.
gauge/gauge_config_policy.hpp holds the lattice_policy and the concrete
gauge_config_policy class, as well as code to instantiate the config_policy from
parameters.

Possible values:
  - HYPERPLANE
  - CYCLE
  - D2h
  - D3h
  - GRID

The sweep policy generalizes the way the phase diagram space is explored.
Previously, gaussian sampling of temperatures as the sole parameter was
hard-coded. The sweep policy is dynamically dispatched, allowing for selection
from parameter file at run time.

Allows to output the individual contraction 'masks' for different ranks. Use
flags `-s | --self` and `-o | --outer` to trigger output of self and outer
contractions, respectively.

No need to build tests in submodules. Speeds up compilation.

Allow for basename with periods (useful with parameter variations and batch
scripting). No longer using the alps::fs functions but relying on C++17's
std::filesystem which is still experimental. Also replaced the ALPS
convenience_params header with a custom version to use those.

Also updated submodules to play nice with C++14.

Support right now for Dinfh (uniaxial) only.

In old ALPSCore versions, there was a race condition in the default ctor of
accumulator_set which meant that the mcbase ctor was not thread-safe. This has
been fixed upstream, such that the awkward workaround (heap alloc of sim class)
is no longer necessary.

data_view class was implemented in a cpp file previously but since the rest of
the SVM project is heavily templated, it's easier to ship only headers that work
with libsvm.[a|so] symbols.