[jetuk]

Allow parts of a model to be made variable, and outputs to be used as
objectives and constraints, for use with external optimisation
algorithms such as multi-objective evolutionary algorithms.

Core:

• VariableParameter trait, implemented by the Constant, Offset,
  MonthlyProfile and RbfProfile parameters, with config validation,
  f64 and u64 values, and bounds.
• The day of year bounds of RbfProfile points are capped at the
  midpoint to their neighbours, so the ranges do not overlap and any
  values within the bounds give each point a unique day of year.
• NetworkVariableConfig and its builder define which parameters are
  variable, in order, and validate each config when it is added.
• Variable values are applied during network setup, before constant
  parameters are computed (Model::setup_with_variables and
  Model::run_with_variables).
• ModelProblem and ModelProblemBuilder combine a model, its
  variables, and objectives and constraints given by the aggregated
  values of recorders. ModelProblem::evaluate runs the model and
  returns the objective and constraint values; it is Send + Sync so
  evaluations can run in parallel.
• Remove the unused per-parameter variable methods from Network.

Schema:

• ProblemConfig defines the variables, objectives and constraints in
  a file separate from the model, and builds a ModelProblem.

Python:

• ProblemConfig, ModelSchema.build_problem, ModelProblem (with
  evaluate and parallel evaluate_batch) and Evaluation, with
  ProblemBuildError and ProblemEvaluateError exceptions.

Docs:

• Add an "Optimisation" page to the book with a random search example.

BREAKING CHANGE: The variable field is removed from the Constant,
Offset and RbfProfile parameters in the schema. Models that set it
will no longer load; define variables in a separate ProblemConfig
instead. RbfProfileVariableSettings is renamed to
RbfProfileVariableConfig, its is_active field is removed, and
value_upper_bounds is now required.