Bug fix in PAC/MCE expectation with cherrypick (was only affecting PAC models with transformed variables).

The RHS of the equation was not acknowledging the user provided name
for the auxiliary variable.

Closes: #1927

Closes: #1721

 make sure user-defined warning states are not overridden

See merge request Dynare/dynare!2285

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

[skip ci]

[skip ci]

[skip ci]

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