Node error UX: reliable ERROR state + visible errors

What happened

Many server node processors ran their work in a bare scope.launch { … } with no try/catch (e.g. ServerTriggerProcessor, ServerDataPointProcessor) or caught exceptions only to log them (the webhook processors). An uncaught exception escaped to the coroutine scope’s CoroutineExceptionHandler, which only logged — so the node never entered NodeState.ERROR, nothing was emitted over SSE, and clients showed a node that looked fine while it had silently failed. Even when an error was recorded, the only place it surfaced was a single editor row gated to EDIT mode.

Fix

• Added launchProcessing(nodeManager, node) { … } (async) and ServerNodeManager.runProcessing(node) { … } (sync) in server/src/jvmMain/kotlin/krill/zone/server/ProcessingScope.kt. Both funnel any uncaught failure into nodeManager.failed(node, …), rethrow CancellationException, and contain any secondary failure raised while reporting (so a captured error can never re-escape).
• Migrated the gap processors to these helpers: ServerTriggerProcessor, ServerDataPointProcessor, ServerLogicGateProcessor, ServerGraphProcessor, ServerWebHookOutboundProcessor, ServerWebHookInboundProcessor, and ServerServerProcessor (which also gained a nodeManager constructor dependency). The webhook outbound processor now acts on the per-method Boolean result it used to discard, so a non-2xx response also reaches ERROR.
• NodeChipScaffold (wrapped by every chip) now reads reactive node state and renders meta.error’s first line, truncated, in colorScheme.error.
• Un-gated ErrorMessageRow() from EDIT mode so the full error shows at the editor bottom across all tabs.

Prevention

• A processor’s process() body must never run in a bare scope.launch or a catch-and-log block. Use launchProcessing / runProcessing so a failure always becomes NodeState.ERROR — the safe path is now the default path.
• New processors are covered by ProcessingScopeTest, which asserts a thrown error reaches ERROR + meta.error and is emitted over the node-update (SSE) pipeline.
• setErrorState no-ops for a node the manager doesn’t yet know, so a failure during a node’s initial registration is still dropped — keep node registration ahead of any fallible processing.
• Remaining bare-scope.launch processors not touched here (Lambda, Pin, Cron, Serial, LLM, Backup, Mqtt, Compute, SMTP, Calculation, TaskList, DataProcessor) should be migrated opportunistically; several already carry correct try/catch + failed(). Audit with: grep -rln "scope.launch" server/src/jvmMain/.../krillapp and check each for a failure path.