refactor(hooks): split useLlm into model / generation / streaming hooks

The 517-line useLlm hook bundled 20 Tauri-command wrappers plus a
single 7-field LlmState object. Consumers (DraftTab and SettingsTab)
used cleanly disjoint subsets: DraftTab touches only streaming +
non-streaming generation (9 functions, 2 state fields), SettingsTab
touches only model lifecycle + context-window config (9 functions, no
state fields). Every streaming token setState therefore re-rendered
SettingsTab, and every model load re-rendered DraftTab, because both
consumers subscribed to a single shared state object.

Splits useLlm into three focused hooks, each owning its own state
slice:

src/hooks/useLlmModel.ts (new)
  State: modelInfo, isLoaded, loading, error
  Commands: checkModelStatus, getLoadedModel, getModelInfo, listModels,
            loadModel, loadCustomModel, validateGgufFile, unloadModel,
            getContextWindow, setContextWindow

src/hooks/useLlmGeneration.ts (new)
  State: generating, error
  Commands: generate, generateWithContext, generateWithContextParams,
            generateFirstResponse, generateChecklist, updateChecklist,
            testModel (+ a runGenerating<T> internal helper that handles
            the common setState/try/finally pattern once instead of
            seven times)

src/hooks/useLlmStreaming.ts (new)
  State: streamingText, isStreaming, generating, error
  Commands: generateStreaming, clearStreamingText, cancelGeneration
  Behavior: keeps the 500KB MAX_STREAMING_TEXT_SIZE truncation contract
            and the UnlistenFn ref-based cleanup

src/hooks/useLlm.ts (rewritten, 517 -> 70 lines)
  Thin orchestrator that composes all three and re-exposes the flat
  API the codebase grew up with. Unifies `generating` (OR of
  generation/streaming) and `error` (first non-null across all three)
  so existing callers and the existing useLlm.test.ts see unchanged
  semantics.

Migrations:
- DraftTab: useLlm() -> useLlmStreaming() + useLlmGeneration(). Drops
  a subscription to loading/isLoaded/modelInfo it never read.
- SettingsTab: useLlm() -> useLlmModel(). Drops a subscription to
  streamingText/isStreaming/generating it never read — the main
  user-visible win of the split, since streaming tokens no longer
  re-render the Settings view when it's open.
- SettingsTab.test.tsx: updates the corresponding vi.mock path from
  "../../hooks/useLlm" to "../../hooks/useLlmModel".

No behavior change. The useLlm orchestrator test (4 cases covering
model mgmt / generation / streaming / errors) still passes unchanged
because the orchestrator's API surface is identical. Typecheck +
vitest (241/241) + eslint clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: saagar210

src/components/Draft/DraftTab.tsx

@@ -26,7 +26,8 @@ import { WorkspaceDialogs } from "./WorkspaceDialogs";
 import { WorkspaceModeShell } from "./WorkspaceModeShell";
 import { WorkspacePanels } from "./WorkspacePanels";
 import { WorkspaceWorkflowStrip } from "./WorkspaceWorkflowStrip";
-import { useLlm } from "../../hooks/useLlm";
+import { useLlmGeneration } from "../../hooks/useLlmGeneration";
+import { useLlmStreaming } from "../../hooks/useLlmStreaming";
 import { useDrafts } from "../../hooks/useDrafts";
 import { useKb } from "../../hooks/useKb";
 import { useAnalytics } from "../../hooks/useAnalytics";
@@ -190,11 +191,13 @@ export const DraftTab = forwardRef<DraftTabHandle, DraftTabProps>(
       isStreaming,
       clearStreamingText,
       cancelGeneration,
+    } = useLlmStreaming();
+    const {
       generateFirstResponse,
       generateChecklist,
       updateChecklist,
       generateWithContextParams,
-    } = useLlm();
+    } = useLlmGeneration();
     const {
       saveDraft,
       updateDraft,

src/components/Settings/SettingsTab.test.tsx

@@ -121,8 +121,8 @@ vi.mock("../shared/Button", () => ({
   ),
 }));
 
-vi.mock("../../hooks/useLlm", () => ({
-  useLlm: () => ({
+vi.mock("../../hooks/useLlmModel", () => ({
+  useLlmModel: () => ({
     loadModel: loadModelMock,
     unloadModel: unloadModelMock,
     getLoadedModel: getLoadedModelMock,

src/components/Settings/SettingsTab.tsx

@@ -14,7 +14,7 @@ import { useDownload } from "../../hooks/useDownload";
 import { useEmbedding } from "../../hooks/useEmbedding";
 import { useJira } from "../../hooks/useJira";
 import { useKb } from "../../hooks/useKb";
-import { useLlm } from "../../hooks/useLlm";
+import { useLlmModel } from "../../hooks/useLlmModel";
 import { useSearchApiEmbedding } from "../../hooks/useSearchApiEmbedding";
 import { useSettingsOps } from "../../hooks/useSettingsOps";
 import {
@@ -72,7 +72,7 @@ export function SettingsTab() {
     setContextWindow,
     loadCustomModel,
     validateGgufFile,
-  } = useLlm();
+  } = useLlmModel();
   const {
     setKbFolder,
     getKbFolder,