feat: Phase 0 foundation — Job entity, expanded pipeline, AI service lockdown, DateApplied history

Unblocks the documented core workflow and closes the AI-service exposure,
without changing existing behaviour.

Job/JobApplication split (additive; see ADR-002):
- New Job entity (the opportunity) with owner-scoped query filter; nullable
  JobApplication.JobId FK. Nothing reads Job yet.
- Migration AddJobEntityAndProspectStages, hand-edited to drop reconciler-owned
  tables the scaffolder re-emitted; verified against the real dev DB.

Pipeline: 10 internal stages across three concerns kept separate —
PipelineStage (workflow) / PipelineGroup (UI: NotApplied/Active/Closed) /
PipelineCategory (analytics). Adds Saved/Interested/Preparing/Withdrawn;
keeps Waiting and Ghosted. Kanban shows 3 grouped columns; cards keep a stage
chip and full transitions; drag applies only safe transitions (never infers
Ghosted/Withdrawn).

DateApplied nullable + SavedAt. Cleared when leaving Applied so analytics stay
accurate; the discarded date is preserved as an AppliedDateCleared JobEvent.

AI service lockdown: no host port; private ai_internal network (backend is the
only other member); X-Ai-Service-Token required on all non-/health endpoints;
AI_SERVICE_TOKEN mandatory via compose. Verified backend-only against the live
stack.

Also carries two pre-existing working-tree files (views/ProfilePage.tsx,
views/CareerWorkspacePage.tsx) so the tree is clean for the branch integration.

Tests: +40 backend (247 total), +5 sidecar (16), +15 frontend.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

tools/summarizer/app.py

@@ -1,4 +1,5 @@
-from fastapi import FastAPI, File, HTTPException, UploadFile
+from fastapi import FastAPI, File, HTTPException, Request, UploadFile
+from fastapi.responses import JSONResponse
 from pydantic import BaseModel, Field
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 from cachetools import TTLCache
@@ -7,6 +8,7 @@ from pypdf import PdfReader
 from docx import Document
 import fitz
 import hashlib
+import hmac
 import io
 import json
 import os
@@ -18,6 +20,33 @@ from urllib.error import URLError, HTTPError
 
 app = FastAPI(title="Local AI Service")
 
+# Shared secret for backend -> ai-service calls. This service has no user auth and can
+# generate against a paid provider (gemini/groq), so an unauthenticated caller on the
+# shared docker network could drain the API key. The port is no longer published to the
+# host (compose uses `expose`), and this header is the second layer.
+#
+# Unset => open, so local dev and the test suite work keyless. Production cannot reach
+# that state: docker-compose declares AI_SERVICE_TOKEN with `:?` so the stack refuses to
+# start without it.
+AI_SERVICE_TOKEN = os.getenv("AI_SERVICE_TOKEN", "").strip()
+AI_SERVICE_TOKEN_HEADER = "X-Ai-Service-Token"
+# /health stays open: the backend probe and the compose healthcheck both call it, and it
+# exposes no user data and no generation path.
+AI_SERVICE_OPEN_PATHS = {"/health"}
+
+
+@app.middleware("http")
+async def require_service_token(request: Request, call_next):
+    if AI_SERVICE_TOKEN and request.url.path not in AI_SERVICE_OPEN_PATHS:
+        supplied = request.headers.get(AI_SERVICE_TOKEN_HEADER, "")
+        # compare_digest to avoid leaking the token through response timing.
+        if not hmac.compare_digest(supplied, AI_SERVICE_TOKEN):
+            return JSONResponse(
+                {"detail": "Invalid or missing service token."},
+                status_code=401,
+            )
+    return await call_next(request)
+
 MODEL_NAME = "sshleifer/distilbart-cnn-12-6"
 MAX_INPUT_CHARS = 20000
 MAX_CONTEXT_CHARS = 2200