Flexibilization of storage of intermediate results in EigenLinearSolver solves memory

Up to now the results of EigenLinearSolver::compute() are always kept and can be reused under some conditions. In the suggested changes the caching is done only when the user enforces this via the tag <linear_solver_compute_only_upon_timestep_change>true</linear_solver_compute_only_upon_timestep_change> in the project file and the time step increment didn't change.

1. Feature description was added to the changelog
2. Tests covering your feature were added? - no new tests
3. Any new feature or behaviour change was documented? - no new features implemented

memory measurements (used examples are described here and here)

process	measurement	current master	configurable linear solver cache
LiquidFlow	peak memory consumption [GB]	4.5	4.2
	Simulation::initializeDataStructures [GB]	3.1	3.1
	Simulation::executeSimulation [GB]	1.4	1.1
HT	peak memory consumption [GB]	11.1	9.8
	Simulation::initializeDataStructures [GB]	5.5	5.5
	Simulation::executeSimulation [GB]	5.6	4.3
ComponentTransport	peak memory consumption [GB]	11.4	10.1
	Simulation::initializeDataStructures [GB]	5.8	5.8
	Simulation::executeSimulation [GB]	5.6	4.3

For those examples the caching mechanism in linear solver wasn't activated and memory could saved.

Run time comparison for benchmarks using the caching

benchmark	run time [s] - master	run time [s] - branch with configurable linear solver cache
ComponentTransport/ReactiveTransport/DecayChain/GlobalImplicitApproach/linear_compute_only_on_dt_change/1d_decay_chain_GIA	2.04	2.08
ComponentTransport/ReactiveTransport/DecayChain/GlobalImplicitApproach/SparseLU_linear_compute_only_on_dt_change/1d_decay_chain_GIA	1.98	1.96
ComponentTransport/ReactiveTransport/DecayChain/GlobalImplicitApproach/varying_dt_linear_compute_only_on_dt_change/1d_decay_chain_GIA	2.24	2.55
1D_HeatConduction_dirichlet_linear_dt_change	0.06	0.07
1D_HeatConduction_dirichlet_varying_dt_linear_compute_only_on_dt_change	0.07	0.07