feat(journal): backend services (moments, search, prep, scheduler, pipeline)

- services/moments.py — CRUD with embedding sync + entity-link helpers
- services/journal_search.py — three-mode search (temporal / entity / semantic)
  with notes-RAG isolation; cosine-similarity helper unit-tested
- services/journal_prep.py — gathers tasks/events/weather/news/projects/
  recent-moments/open-threads into a structured prep block
- services/journal_scheduler.py — per-user APScheduler cron for daily prep,
  follows the BackgroundScheduler + threadsafe-async pattern from event_scheduler
- services/journal_pipeline.py — system prompt (persona + calibration rules)
  with last-48h ambient moments injection
- app.py: wire journal scheduler start/stop hooks
- routes/settings.py: re-add live-reschedule on timezone change (now via
  journal_scheduler.update_user_schedule)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/fabledassistant/services/journal_pipeline.py

@@ -0,0 +1,117 @@
+"""System prompt + ambient context injection for journal conversations.
+
+Distinct from the chat pipeline (services/llm.py) in three ways:
+1. New persona: warm, curious listener; not a task manager.
+2. Calibration rules: ask before structural changes; record_moment freely.
+3. Auto-injects last ~48h of moments for ambient continuity (no notes-RAG).
+
+Notes-RAG auto-injection MUST be disabled for journal sessions — the
+journal's ambient context replaces it. Failing to disable would let notes
+leak into journal sessions, violating the design's isolation invariant.
+"""
+from __future__ import annotations
+
+import datetime
+
+from fabledassistant.services.journal_search import search_journal
+from fabledassistant.services.user_profile import build_profile_context
+
+JOURNAL_PERSONA = (
+    "You are a warm, curious listener helping the user record their day. "
+    "You are NOT a task manager. Your role is to listen, ask gentle "
+    "follow-up questions when something seems significant or underspecified, "
+    "and quietly maintain structure as a byproduct of conversation."
+)
+
+JOURNAL_CALIBRATION = """\
+CALIBRATION (very important):
+- When the user mentions a person you don't know, ask: "Is <Name> someone I should remember?" Confirm before calling update_person.
+- When a person reference is ambiguous (multiple matches), ask which one. Never guess.
+- When the user says something action-implying (e.g., "I finished X"), ASK before calling update_task. Silent updates feel derailing.
+- Use the existing tool's `confirmed=false` -> user-confirms -> `confirmed=true` pattern. The frontend renders the inline confirm UI automatically.
+- record_moment is the EXCEPTION: call it freely and silently when the user mentions a meaningful beat. Moments are cheap and user-correctable in the timeline view; gating them would kill the flow.
+- Do NOT call set_rag_scope. The journal scope is implicit.
+- Notes are not auto-retrieved here. If you need to reference notes, call search_notes explicitly.
+"""
+
+PHASE_GREETINGS = {
+    "morning": "Morning. What's the day looking like for you?",
+    "midday": "How's it going so far?",
+    "evening": "Wrapping up — how'd the day shake out?",
+}
+
+
+def determine_phase(
+    *,
+    now_local: datetime.datetime,
+    day_rollover_hour: int = 4,
+    morning_end_hour: int = 12,
+    midday_end_hour: int = 18,
+) -> str:
+    """Return 'morning' | 'midday' | 'evening' for a local datetime."""
+    h = now_local.hour
+    if h < day_rollover_hour:
+        return "evening"
+    if h < morning_end_hour:
+        return "morning"
+    if h < midday_end_hour:
+        return "midday"
+    return "evening"
+
+
+def phase_greeting(phase: str) -> str:
+    return PHASE_GREETINGS.get(phase, PHASE_GREETINGS["morning"])
+
+
+async def build_journal_system_prompt(
+    *,
+    user_id: int,
+    day_date: datetime.date,
+    user_timezone: str,
+) -> str:
+    """Static-then-dynamic system prompt.
+
+    Static prefix (persona + calibration) is identical on every request,
+    preserving Ollama KV cache. Dynamic suffix changes per-day.
+    """
+    static_block = f"{JOURNAL_PERSONA}\n\n{JOURNAL_CALIBRATION}"
+
+    today_iso = day_date.isoformat()
+    tz_block = f"Today is {today_iso} ({user_timezone})."
+
+    profile_context = await build_profile_context(user_id)
+    profile_section = f"\n\n{profile_context}" if profile_context else ""
+
+    ambient = await _ambient_moments_block(user_id=user_id, day_date=day_date)
+    ambient_section = (
+        f"\n\nRECENT JOURNAL CONTEXT (last 48h):\n{ambient}" if ambient else ""
+    )
+
+    return f"{static_block}\n\n{tz_block}{profile_section}{ambient_section}"
+
+
+async def _ambient_moments_block(
+    *, user_id: int, day_date: datetime.date
+) -> str:
+    """Render last 48h of moments as a compact text block.
+
+    Capped at 20 moments / 1500 chars total. Distinct from RAG retrieval.
+    """
+    moments = await search_journal(
+        user_id=user_id,
+        date_from=day_date - datetime.timedelta(days=2),
+        date_to=day_date,
+        limit=20,
+    )
+    if not moments:
+        return ""
+
+    lines: list[str] = []
+    total_chars = 0
+    for m in moments:
+        line = f"- [{m['occurred_at']}] {m['content']}"
+        if total_chars + len(line) > 1500:
+            break
+        lines.append(line)
+        total_chars += len(line)
+    return "\n".join(lines)