Build-loop type-stability: branch-divergent PTDF balance values and AbstractArray staging dict

Summary

Two build-loop type-stability findings from a time-series-chain audit. Both compound the container type-erasure documented in Sienna-Platform/InfrastructureOptimizationModels.jl#155 — fixing them without that one helps little, so they're documented here for when that issue is addressed.

Findings

1. Branch-divergent per-timestep values in the PTDF balance expression builders. src/common_models/add_to_expression.jl:1533-1578 (branch at 1565-1572) and _add_load_ts_parameter_to_balance! (:224-256, branch at 246-255): param/mult are either a JuMP scalar from get_parameter_column_refs(...)[t] or a literal Float64, chosen by a runtime has_ts flag resolved per device but not per type. In isolation a fine 2-way union; through the erased containers of IOM#155 it collapses to Any, so add_proportional_to_jump_expression! (concrete dispatch arms in add_jump_expressions.jl:24-43) redispatches dynamically 2 × devices × timesteps. Fix direction: resolve has_ts into a Val{true}/Val{false} split (or two typed helpers called once per device) so the timestep body is monomorphic.

2. Dict{String, AbstractArray} staging in the time-series parameter build. src/common_models/add_parameters.jl:179 (initial_values = Dict{String, AbstractArray}()) erases the concrete Vector{Float64} the accessors return; the values re-emerge as ::AbstractArray, the unwrap_for_param.(...) broadcast widens to Any eltype, and IOM._set_parameter_at! (5 concrete dispatch arms) redispatches per step in the loop at :257-259. Mirror pattern in _add_objective_function_parameters! (~:727-782). Bounded today — runs once per build!, and no update_container_parameter_values! implementation exists yet on this branch to re-execute it per step — but it becomes hot the moment that update path is ported. Fix: type the dict to the concrete family (Dict{String, Vector{Float64}}), a signature-only change.

Found during a type-stability audit (Julia 1.12.5, faithful minimal reproductions; _set_parameter_at! and the multiplier dispatch tables themselves verified clean — the callers hand them Any).

Start with the cited branches in src/common_models/add_to_expression.jl and the concrete dispatch arms in add_jump_expressions.jl, then inspect src/common_models/add_parameters.jl:179 and the mirrored objective-parameter code around lines 727-782. Use the stated Julia 1.12.5 minimal reproductions to verify the callers no longer hand Any values to the dispatch tables; done means the timestep paths remain type-stable without changing their behavior, alongside the container-type work tracked in IOM#155.