fix: backshift additional_observed when generating equations

[AayushSabharwal]

No description provided.

[oscardssmith]

Can this get a comment explaining the purpose (and ideally a test?)

[AayushSabharwal]

The additional_observed field of TearingState contains observed equations identified through processes other than tearing. Typically, this was only populated by the trivial_tearing! pass in StateSelection. However, MTK handles discrete systems by shifting them forward by a step (x(k) becomes x(k+1)). It omitted doing so on ::TearingState.original_eqs, which is what trivial_tearing! relies on. As a result, trivial_tearing! never did anything on discrete systems (since the equations didn't match what it expected) and additional_observed was never tested for such systems. SciML/ModelingToolkit.jl#4402 utilizes additional_observed for all systems, which exposed this bug. This PR now appropriately shifts original_eqs, which means trivial_tearing! does its job (also part of the test) and additional_observed is backshifted in reassembly to generate appropriate observed equations.

[AayushSabharwal]

Minor bug in the test. I'll let CI run, and if everything else passes push the fix and merge once that one job completes.