feat(inspect): port legacy parser rules the new resolver missed (installed nodes)

R6 and R9 for nodes whose sources are installed beside ComfyUI, each
entry's input names and output slots read from the node's code (the task
report lists file and line):

- detailer stages UltimateSDUpscale (image), UltimateSDUpscaleNoUpscale
  (upscaled_image) and DetailerForEach (image), judged with their upstream
  samplers like FaceDetailer;
- pass-throughs ControlNetInpaintingAliMamaApply and LTXVConditioning
  (positive 0, negative 1) and FluxKontextMultiReferenceLatentMethod;
- encoder Text to Conditioning, strings TextInput and DisplayText;
- facts model pass-throughs Lora Loader (LoraManager), Power Lora Loader
  (rgthree), ModelPatchTorchSettings, CFGNorm, PerturbedAttentionGuidance.

Impact Pack encodes a detailer's non-empty wildcard instead of its positive
input, so FaceDetailer and DetailerForEach now list it as a prompt
override: a literal wildcard is decided at run time and a linked one names
its source, instead of resolving to the positive input. The overwritten
PromptManager refusal (R1) still holds behind the new pass-throughs. The
default profile does not read the table.

tests/js/inspection-legacy-ports.test.js

@@ -57,3 +57,42 @@ test("R10: commas inside quoted settings values do not split them", () => {
     // An unbalanced quote is not a quoted value: the historical split, as A1111 reads it.
     assert.deepEqual(parameterFacts(`a cat\n${FOOTER}, Note: "open, Model: base`).model, ["known", "base"]);
 });
+
+function graphFacts(graph, resolution) {
+    const facts = extractGenerationFacts({
+        prompt_graph: { status: "parsed", value: graph },
+        workflow_graph: { status: "absent", value: null },
+        metadataStatus: "available",
+        resolution,
+    });
+    return Object.fromEntries(fields.map(field => [field, [facts[field].status, facts[field].value]]));
+}
+
+function modelGraph(between) {
+    return {
+        ckpt: { class_type: "CheckpointLoaderSimple", inputs: { ckpt_name: "base.safetensors" } },
+        hop: { class_type: between, inputs: { model: ["ckpt", 0], clip: ["ckpt", 1] } },
+        s: { class_type: "KSampler", inputs: { seed: 5, steps: 20, cfg: 7, sampler_name: "euler", scheduler: "normal", model: ["hop", 0] } },
+    };
+}
+
+test("R9: model pass-throughs verified in their sources lead to the loader", () => {
+    for (const name of ["Lora Loader (LoraManager)", "Power Lora Loader (rgthree)", "ModelPatchTorchSettings", "CFGNorm", "PerturbedAttentionGuidance"]) {
+        assert.deepEqual(graphFacts(modelGraph(name)).model, ["known", "base.safetensors"], name);
+        // Only output 0 is the model.
+        const other = modelGraph(name); other.s.inputs.model = ["hop", 1];
+        assert.deepEqual(graphFacts(other).model, ["unresolved", null], name);
+    }
+    assert.deepEqual(graphFacts(modelGraph("ModelMergeSDXL")).model, ["unresolved", null]);
+});
+
+test("R6: upscale and SEGS detailers supply stage facts like FaceDetailer", () => {
+    for (const [class_type, image] of [["UltimateSDUpscale", "image"], ["UltimateSDUpscaleNoUpscale", "upscaled_image"], ["DetailerForEach", "image"]]) {
+        const graph = modelGraph("CFGNorm");
+        graph.decode = { class_type: "VAEDecode", inputs: { samples: ["s", 0], vae: ["ckpt", 2] } };
+        graph.d = { class_type, inputs: { [image]: ["decode", 0], model: ["hop", 0], seed: 7, steps: 12, cfg: 7, sampler_name: "euler", scheduler: "normal" } };
+        assert.deepEqual(graphFacts(graph, { state: "resolved", sampler_id: "d" }).seed, ["known", "7"], class_type);
+        const merged = graphFacts(graph, { state: "ambiguous", sampler_id: "d" });
+        assert.deepEqual([merged.seed[0], merged.steps[0], merged.cfg], ["ambiguous", "ambiguous", ["known", "7"]], class_type);
+    }
+});