feat(diarizer/nemotron3): load monolithic presets from monolithic/v2 (M3 ANE compile fix)

Sources/FluidAudio/Diarizer/Nemotron3/Nemotron3Models.swift

@@ -137,7 +137,7 @@ public struct Nemotron3Models {
     /// (if not cached) and load it.
     ///
     /// Layout under `cacheDirectory` (default `~/Library/Application Support/FluidAudio/Models`):
-    /// `nemotron-3-diarization/{monolithic,split}/<bundle>.mlmodelc` plus the root `.bin` assets.
+    /// `nemotron-3-diarization/{monolithic/v2,split}/<bundle>.mlmodelc` plus the root `.bin` assets.
     /// Only the requested preset's bundle is fetched.
     public static func loadFromHuggingFace(
         config: Nemotron3Config,

Sources/FluidAudio/Diarizer/Nemotron3/Nemotron3Types.swift

@@ -82,7 +82,11 @@ public struct Nemotron3Config: Sendable {
     }
 
     /// Subdirectory of `FluidInference/nemotron-3-diarization-coreml` holding this preset's bundle.
-    public var hubSubdirectory: String { splitGraph ? "split" : "monolithic" }
+    ///
+    /// Monolithic presets load from `monolithic/v2/`: fp16-I/O, gather-free re-exports that
+    /// compile for the ANE on M3-class chips (#951). The originals stay in `monolithic/` for
+    /// older releases.
+    public var hubSubdirectory: String { splitGraph ? "split" : "monolithic/v2" }
 
     // MARK: Presets (model card recommended profiles)
 

Tests/FluidAudioTests/Diarizer/Nemotron3/Nemotron3CacheVersionTests.swift

@@ -79,4 +79,11 @@ final class Nemotron3CacheVersionTests: XCTestCase {
         let absent = root.appendingPathComponent("does-not-exist")
         XCTAssertNoThrow(try Nemotron3Models.discardStaleCache(at: absent))
     }
+
+    func testMonolithicPresetsLoadFromV2AndSplitPresetsFromSplit() {
+        for config in [Nemotron3Config.low, .offline, .fast, .fast32, .fast128] {
+            XCTAssertEqual(config.hubSubdirectory, "monolithic/v2", config.modelFileName)
+        }
+        XCTAssertEqual(Nemotron3Config.preset(named: "fast32-split-w8a8")?.hubSubdirectory, "split")
+    }
 }