Fill event models from the emulator's system at simulation build

[m-bossart]

Draft until Sienna-Platform/PowerOperationsModels.jl#404 merges. This PR uses it, and [sources] pulls POM from main, so CI fails until then.

Fixes #1701.

Each model's build filled the shared event models from its own system.
The outages the emulator applied therefore depended on build order, and their attribute ids could belong to a decision model's system, so the emulator could crash or apply the wrong outage.

The emulator's system is now the single source of truth for which outages exist:

• Before any model builds, the simulation fills every event model from the emulator's system with POM's discover_event_attribute_device_map.
  Each model then applies the event to the outaged devices it has.
• A sequence with events and no emulation model fails the build with IS.ConflictingInputsError.
• apply_simulation_events! looks attributes up in the emulator's system by id, as before, and fails loudly if one is not of the event model's type.
• Outage attributes in a decision model's own system do not affect events.
  They may serve other formulations, such as security-constrained N-1 in a UC → RT → emulator setup.

An outaged device no longer has to be in any decision model:

• Event-parameter state starts at its no-outage values (countdown and offsets 0, availability 1) instead of NaN, since the emulator reads its event parameters from decision state before any event step writes them.
• Event-parameter keys only the emulator has get a decision state of their own.
• The writers that reset an outaged device's variables (OnVariable, ActivePowerVariable, TimeDurationOn/Off) skip a device whose variable state no model writes.

Tests (test/test_events.jl):

• The diverging-id reproduction from Regression on psy6: simulation events look up outages by attribute id across systems #1701 runs to completion with the expected outages.
• The emulator outages Alta and Park City while D1's system carries its own Brighton outage: both decision models get event parameters for exactly Alta and Park City, and D2 sees Park City unavailable with zero dispatch during the outage.
• An outage on a device only the emulator has is applied by the emulator and leaves the decision models alone, while a shared device's outage still reaches them; the same holds when no decision model has event parameters for the device type.
• A sequence with events and no emulator fails the build.
• With rebuild_model = true on every model, the simulation runs to completion.

The full suite passes except test_market_bid_cost.jl:388 (2 failures), which also fail without this change.

🤖 Generated with Claude Code