Solver capabilities

BayesOptAlg declares its capabilities through solver_capabilities:

• linear_eq = true, nonlinear_eq = true: expensive equality constraints are transformed into two inequality constraints (of any linearity),
• linear_ineq = true, nonlinear_ineq = true,
• box = true,
• continuous = true: the surrogate machinery models variables as continuous,
• the semidefinite constraints (linear_sd, nonlinear_sd), the discrete variable types (binary, integer) and infinite_bounds are forwarded from the sub-solver, since the surrogate machinery passes those model features through untouched.

For example BayesOptAlg(IpoptAlg()) inherits linear_sd = false, nonlinear_sd = false, binary = false, integer = false and infinite_bounds = true from Ipopt's declaration, so models with semidefinite constraints or integer variables error with an informative message before optimization starts. The surrogate subproblem is validated against the sub-solver's declaration via checksubproblem!.