warning_config.m should only be called once in dynare.m

The initial guess was set to the solution, so the algorithms would really be
tested.

Function Interpreter::check_for_controlled_exo_validity() would give incorrect
results because the sconstrained_extended_path structure was containing 1-based
variable indices (it is expecting 0-based ones).

This prevents a name clash with some function-local variables also named “b”.

manual: clarify treatment of missing data in Inversion Filter

See merge request Dynare/dynare!2284

Replaced

         ᵐ                              ᵐ⁻¹ ᵐ⁻¹
  Zₜ = - ∑  𝛼ᵢ 𝛽ⁱ⁺¹ Zₜ₊ᵢ + A(1) [ 𝛥 yₜ + ∑   ∑   𝛼ⱼ₊₁𝛽ʲ⁺¹𝛥 yₜ₊ₖ ]
        ᵢ₌₁                             ₖ₌₁ ⱼ₌ₖ

by

         ᵐ                              ᵐ⁻¹ ᵐ⁻¹
  Zₜ = - ∑  𝛼ᵢ 𝛽ⁱ⁺¹ Zₜ₊ᵢ + A(1) [ 𝛥 yₜ - ∑   ∑   𝛼ⱼ₊₁𝛽ʲ⁺¹𝛥 yₜ₊ₖ ]
        ᵢ₌₁                             ₖ₌₁ ⱼ₌ₖ

Also:

 - Added comments and comparison with the formula in Brayton et alii (2000),
 - Removed unused input argument,
 - Removed unused variables.

Was crashing if the PAC target was composite.

NK_baseline_steadystate.m: cosmetic changes

See merge request Dynare/dynare!2282

 Fix marginal linearization in the context of perfect_foresight_with_expectation_errors_solver with homotopy

The formula used to create the scenario in subsequent informational periods was
wrong.

See also message of commit c3d91d5c.

 ver_less_than.m: account for equality of versions

See merge request Dynare/dynare!2281

Ref. !2279

Ref. !2280

Ref. !2277

 univariate_kalman_filter: discard pathological parameter draws that result in negative variance

Logic was faulty as not bigger than tolerance does not imply being 0