fix(feedback): Phase 2 follow-up — Codex review MAJORs + NIT

Independent review surfaced three majors and one nit. All actioned.

MAJOR 1 — profile scoping:
`decompose_and_store` and `extract_tasks_from_transcript_cmd` now
accept an optional `profile_id` (wired from `profilesStore.activeProfileId`
in MicroSteps.svelte and DictationPage.svelte), and thread it into the
feedback-retrieval query so per-profile decomposition styles do not
leak into each other. `record_feedback` gets the same treatment.

MAJOR 2 — prompt-budget regression on long inputs:
New `trim_to_budget` helper + `FEW_SHOT_CHAR_BUDGET = 2000` char cap
in `src-tauri/src/commands/tasks.rs`. Retrieval still pulls up to 5
rows but they are char-counted and truncated against the budget
before being sent to the LLM. Char cost matches the `Input: ...\n
Good output: ...` render path so the budget maps cleanly to ~570
Qwen3 tokens, well inside the 8192-context reserve after the 512-
or 768-token response allocation. Oldest-first drop order (iteration
stops at cost exceeded) preserves the most recent correction which
is the one carrying the user's live preference.

MAJOR 3 — inline edit stale-rollback race:
`saveEdit` in MicroSteps.svelte now stamps a monotonic per-step
`saveToken`. Each edit bumps the token; on failure the rollback
only fires if `saveToken[step.id] === myToken`, so a slow-failing
first save can no longer overwrite a faster successful second save.

NIT — retrieval ordering stability:
`list_feedback_examples` ORDER BY now `created_at DESC, id DESC`.
SQLite timestamp precision is one second; without the secondary
key, bursty feedback within the same second would select
non-deterministically.

Also: malformed `context_json` now warns via eprintln! rather than
disappearing silently — Codex minor.

All green: cargo build + 249 tests + clippy -D warnings + fmt
+ svelte-check (0/0) + npm run build.

src/lib/components/MicroSteps.svelte

@@ -1,6 +1,7 @@
 <script lang="ts">
   import { invoke } from '@tauri-apps/api/core';
   import { ListTree, Check, Timer, Loader2, ThumbsUp, ThumbsDown, Pencil } from 'lucide-svelte';
+  import { profilesStore } from '$lib/stores/profiles.svelte.ts';
 
   let { parentTaskId, parentTaskText = '', reduceMotion = false } = $props();
 
@@ -24,6 +25,13 @@
   let editing = $state<Record<string, boolean>>({});
   let draft = $state<Record<string, string>>({});
 
+  // Monotonic token per-step. Each time the user kicks off a save we
+  // bump the token; the pending save remembers its token and only
+  // rolls back if the failure belongs to the still-current edit.
+  // Without this, a slow first save that eventually fails will stomp
+  // a faster second save that already committed.
+  let saveToken = $state<Record<string, number>>({});
+
   async function loadSubtasks() {
     loading = true;
     error = '';
@@ -40,7 +48,10 @@
     decomposing = true;
     error = '';
     try {
-      subtasks = await invoke<Subtask[]>('decompose_and_store', { parentTaskId });
+      subtasks = await invoke<Subtask[]>('decompose_and_store', {
+        parentTaskId,
+        profileId: profilesStore.activeProfileId,
+      });
     } catch (e) {
       error = String(e);
     } finally {
@@ -94,6 +105,7 @@
           originalText: step.text,
           correctedText: null,
           contextJson: feedbackContextJson(),
+          profileId: profilesStore.activeProfileId,
         },
       });
     } catch (_) { /* feedback capture is best-effort, never fatal */ }
@@ -116,6 +128,10 @@
     const original = step.text;
     // Update in-memory first so the UI is snappy; roll back if the
     // persistence call fails so we never show stale-but-different text.
+    // The monotonic `myToken` guards against a stale rollback stomping
+    // a fresher successful save that landed while we were waiting.
+    const myToken = (saveToken[step.id] ?? 0) + 1;
+    saveToken = { ...saveToken, [step.id]: myToken };
     const idx = subtasks.findIndex(s => s.id === step.id);
     if (idx >= 0) subtasks[idx] = { ...subtasks[idx], text: next };
     try {
@@ -132,10 +148,18 @@
           originalText: original,
           correctedText: next,
           contextJson: feedbackContextJson(),
+          profileId: profilesStore.activeProfileId,
         },
       }).catch(() => {});
     } catch (_) {
-      if (idx >= 0) subtasks[idx] = { ...subtasks[idx], text: original };
+      // Only roll back if our token is still the most recent — a later
+      // edit that already succeeded must not be overwritten.
+      if (saveToken[step.id] === myToken) {
+        const rollbackIdx = subtasks.findIndex(s => s.id === step.id);
+        if (rollbackIdx >= 0) {
+          subtasks[rollbackIdx] = { ...subtasks[rollbackIdx], text: original };
+        }
+      }
     }
   }
 

src/lib/pages/DictationPage.svelte

@@ -487,7 +487,10 @@
     if (settings.aiTier === "tasks" && llmLoaded) {
       markGenerating("Extracting tasks");
       try {
-        const items = await invoke("extract_tasks_from_transcript_cmd", { transcript: text });
+        const items = await invoke("extract_tasks_from_transcript_cmd", {
+          transcript: text,
+          profileId: profilesStore.activeProfileId,
+        });
         markGenerationDone(true);
         return items.map((taskText) => ({ text: taskText }));
       } catch (err) {