Name the component and key when a simulation state value was never written

[m-bossart]

Summary

Simulation state now covers the union of each key's columns across models (3a1dabe).
A component that no model writes stays NaN for the whole run.
When a model read that value, the failure was unclear:

• Initial-condition updates hit IOM's @assert isfinite in set_ic_quantity!, which prints the whole component and names neither the state key nor the cause.
• The emulator feedforward update copied the value into the model without any check.

On the previous code, the same setup failed at build with a message about device filters.

Changes

• _check_state_value (src/simulation/simulation_state.jl) raises IS.InvalidValue naming the component, the state key, and both causes: a decision not yet made for the time being read, or no model writing that component, usually from different device filters.
• Every initial-condition update reads through _get_state_value, which applies the check.
• The five copies of the finite-value error in update_container_parameter_values.jl call the shared check.
• The emulator feedforward update (VariableValueAttributes, EmulationModel) now checks too.
• The decision-model event parameter update builds its ParameterKey once per call and uses it for its state reads and the check.

The check runs only on values a model actually reads.
Unwritten cells for components nobody reads are expected under union columns and are not flagged.

Testing

• New end-to-end test in test/test_simulation_state_union_columns.jl: an emulator whose UC formulation filters out Alta leaves Alta's OnVariable state unwritten, and UC's initial conditions read it at step 2. The test checks that the simulation log names Alta, OnVariable__ThermalStandard, and device filters. It fails without the change.
• Full suite against current IS4, IOM main and POM main: 29,715 pass, 2 fail.
  The 2 failures are the hot-start coverage check at test/test_market_bid_cost.jl:388, which fails the same way on psy6. POM now reads time limits in minutes, and that test's multi-start scenarios no longer produce a hot start. That fix is a separate test-data change.

🤖 Generated with Claude Code