Add a solver characteristic for solvers implementing solve_from()

[nhuet]

The new characteristic, the generic (default) version and its optional specific declination are:

• generic characteristic: FromInitState
• specific characteristic: FromAnyState:
  • the solver have a solve_from() method
  • solve() use it by default to solve from initial state (if domain is initializable else raise a ValueError if not implemented)
  • solve() (and _solve) gets also an optional argument from_memory to solve from another state (and call solve_from accordingly)

Consequences:

• add from_memory arg in Domain.solve_with()
• Solver inherits from FromInitState (to have solve() and _solve() in its
  api)
• FromAnyState inherits from FromInitState and overrides solve() and _solve()
• init_solve() added to FromAnyState to be called in _solve(), to ensure solve_from() will already have the domain (and for instance the underlying c++ solver) initialized when called from _solve.

Minor modifications to avoid circular imports

• We import Domain for annotations only when type checking
• We use a dedicated logger in parallel_domains.py to avoid havind to import domains

We adapt all solvers to this new api:

• for solvers already implementing solve_from():
  • derive from FromAnyState
  • remove _solve implementation (already in FromAnyState)
• for LazyAstar and LRTAstar:
  • remove attribute self._from_state
  • derive from FromAnyState
  • implement _init_solve() to init self._domain
  • convert _solve into _solve_from
• for MAHD:
  • derive FromAnyState, but raises an error if chosen solver cannot do it
• for other solvers: nothing changes.

[fteicht]

Thanks @nhuet !