feat(ui): add Svelte frontend for /humanize, LibreTranslate option for Step 4, and Docker startup fixes

[eyeofastarte]

Summary

This PR bundles the last 3 commits:

• Docker fix so the container actually starts reliably.
• A proper Svelte UI for the /humanize flow, built and served by FastAPI.
• Support for self-hosted LibreTranslate as a Step 4 provider in the v1.5 Standard Pipeline (previously Niutrans-only).

Type of Change

• Bug fix (non-breaking fixes a confirmed issue)
• New feature (non-breaking adds new behaviour)
• Breaking change (changes or removes existing behaviour)
• Refactor / cleanup (behaviour unchanged)
• Documentation only
• CI / tooling / configuration

Why

• The Docker path had dependency/startup issues; the container didn't start up cleanly.
• There was no first-party UI, so trying /humanize meant curl or Postman.
• Step 4 hard-required Niutrans. If you didn't have access you were stuck. LibreTranslate gives a self-hosted, free, testable output.

Changes

Docker / runtime

• docker/Dockerfile now runs the unified entrypoint: uvicorn src.api:app --port 8001.
• docker-compose.yml healthcheck hits http://localhost:8001/v1/health, so it reflects the v1 app instead of just "the port is open."

Frontend (Svelte)

• New app under frontend/src/*: text input, method + language pickers, result card, error states.
• frontend/src/lib/api.js: small API helper that checks status/content-type and throws readable errors when the server returns non-JSON (e.g. a proxy HTML error page) instead of blowing up on res.json().
• Dockerfile is now multi-stage: builds the frontend and copies dist into /app/static, where FastAPI serves it. One container, no separate dev server in prod.

LibreTranslate Step 4 provider (v1.5 Standard Pipeline)

• New libretranslate_translate(...) in src/standard/translators.py.
• src/standard/pipeline.py gets pipeline.step4_provider = "niutrans" | "libretranslate": defaults to Niutrans, so existing setups are unaffected.
• /v1/config (src/standard/api.py) now exposes step4_provider and libretranslate_base_url: non-sensitive values only.
• Exports updated in src/standard/__init__.py.
• config/config.toml documents the new keys: [pipeline] step4_provider and [providers.libretranslate].

Tests

• tests/test_standard_api.py: route shapes/validation, config redaction, language normalization, legacy bridge reachability, and the new /v1/config fields.
• tests/test_standard_pipeline.py: Step 4 defaults to Niutrans, switches to LibreTranslate when configured, and the response shape stays the same either way.

Visual / UI changes - REQUIRED if you touched anything that renders

[molly554]

Review: Changes Requested

Tests pass (38/38), code is well-structured overall, but there are a few issues that need fixing before merge.

---

🔴 Critical: Branch deletes Atlas Cloud provider (#28)

This branch was forked before PR #28 (Atlas Cloud) was merged into main. As a result, merging this PR will remove the Atlas Cloud provider, its env-var fallback, and its 2 tests from llm_client.py and test_llm_client.py.

Fix: Rebase onto current main and resolve conflicts — keep the Atlas Cloud entries.

---

🟡 Breaking: Port change 8000 → 8001

humanizer.py CLI --serve and docker-compose.yml / Dockerfile all switch from port 8000 to 8001 without a migration note. Existing Docker users will break silently.

Fix: Either keep 8000 (preferred — less churn), or document the port change prominently in the PR description and README.

---

🟡 Dependencies: fastapi/uvicorn/pydantic added to base requirements.txt

PR #17 was rejected partly for pulling FastAPI into the base dependency set. The pipeline's core use case is CLI — users who just want python -m src.standard.pipeline shouldn't need FastAPI installed.

Fix: Move fastapi, uvicorn[standard], pydantic to setup.py extras:

extras_require={
    "api": ["fastapi>=0.109.0", "uvicorn[standard]>=0.27.0", "pydantic>=2.0.0"],
    ...
}

---

🟡 API has no authentication

/v1/humanize/standard is open — anyone who can reach the port can burn through your LLM API key. A Docker deployment exposed on a public IP would be a cost risk.

Fix: Add a simple bearer-token check (read from env/config), or at minimum document the risk and recommend running behind a reverse proxy.

---

🟢 Minor issues

1. Unused imports in src/standard/api.py: time and Optional are imported but never used — remove them.

2. _load_config() reads disk on every request: Consider caching with @lru_cache or loading once at startup via FastAPI lifespan. Current approach is fine for low traffic but will add latency under load.

3. normalize_target_lang() is verbose: The 12-language if-chain and the final fallback target_lang.split("-")[0] do the same thing. The fallback alone covers all cases — the explicit branches are unnecessary.

4. SPA catch-all route ordering: @app.get("/{full_path:path}") in humanizer.py catches all GET requests when STATIC_DIR exists. This works now because the legacy API only has POST endpoints, but adding a new GET route later would be silently shadowed. Consider using a more specific prefix like @app.get("/app/{full_path:path}").

---

Summary

The LibreTranslate integration, Svelte UI, and API layer are solid work — good test coverage and clean separation. The main blocker is the rebase conflict that would delete Atlas Cloud. Once that's resolved and the dependency/auth issues are addressed, this is ready to merge.

[eyeofastarte]

Addressed review items: rebased onto latest main, reverted ports to 8000, moved FastAPI stack to optional api extra, added docker-compose security warning for unauthenticated API exposure.