feat: air-gapped deployment guide, no implicit downloads

- Ship tiktoken's cl100k_base inside the package and build the encoding
  from it, so token counting never downloads anything.
- Default EMBEDDINGS_CACHE_DIR to <data home>/models instead of FastEmbed's
  temp dir, and read tokenizer.json and repo metadata from that cache, so
  a model downloads once and survives reboots.
- TTS_PROVIDER=none and STT_PROVIDER=none switch the speech features off:
  the endpoints return 404, audio files fail to ingest with a clear
  message, /api/config reports tts_available/stt_available, and the UI
  hides the Speak and microphone buttons.
- Drop the Google Fonts Roboto import from the web UI.
- prefetch-models fills the cache the app reads; verify-offline checks the
  packaged encoding.
- Docs: new Air-Gapped Deployment guide, settings and cache notes.

docs/content/Deploying/Pip-Install.mdx

@@ -49,7 +49,7 @@ pipx runpip docsgpt install --force-reinstall --no-deps --index-url https://down
 
 ## Configure
 
-DocsGPT keeps its runtime files in a **data home**: the `.env` file it reads settings from, uploaded files under `inputs/` and vector indexes under `indexes/`. The data home is the directory you run the commands from, or the directory `DOCSGPT_HOME` points to. `DOCSGPT_ENV_FILE` points at a `.env` kept somewhere else. Both variables must be set in the process environment, not in `.env`: they decide where `.env` is read from.
+DocsGPT keeps its runtime files in a **data home**: the `.env` file it reads settings from, uploaded files under `inputs/`, vector indexes under `indexes/` and downloaded embedding models under `models/`. The data home is the directory you run the commands from, or the directory `DOCSGPT_HOME` points to. `DOCSGPT_ENV_FILE` points at a `.env` kept somewhere else. Both variables must be set in the process environment, not in `.env`: they decide where `.env` is read from.
 
 Create a `.env` in the data home. The minimum for a hosted LLM:
 
@@ -87,7 +87,7 @@ The worker is not optional: query embedding runs on it, so search fails without
 Other commands:
 
 - `docsgpt api --reload`: a development server with auto-reload.
-- `docsgpt prefetch-models`: download the embedding, tokenizer and parser models ahead of time, for machines that go offline.
+- `docsgpt prefetch-models`: download the embedding models and their tokenizers ahead of time, for machines that go offline (see [Air-Gapped Deployment](/Deploying/Air-Gapped)).
 - `docsgpt verify-offline`: check that a prepared install starts with networking off.
 - `docsgpt reembed`: re-embed every index after changing `EMBEDDINGS_NAME` (see [Upgrading](/upgrading)).