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.

docsgpt/parser/file/audio_parser.py

@@ -2,7 +2,7 @@ from pathlib import Path
 from typing import Dict, Union
 
 from docsgpt.core.settings import settings
-from docsgpt.parser.file.base_parser import BaseParser
+from docsgpt.parser.file.base_parser import BaseParser, DocumentParseError
 from docsgpt.stt.stt_creator import STTCreator
 from docsgpt.stt.upload_limits import enforce_audio_file_size_limit
 
@@ -17,6 +17,10 @@ class AudioParser(BaseParser):
 
     def parse_file(self, file: Path, errors: str = "ignore") -> Union[str, list[str]]:
         _ = errors
+        if not STTCreator.is_enabled(settings.STT_PROVIDER):
+            raise DocumentParseError(
+                f"{file.name}: audio files need speech-to-text, which is disabled (STT_PROVIDER=none)."
+            )
         try:
             enforce_audio_file_size_limit(file.stat().st_size)
         except OSError:

docsgpt/parser/tokenization.py

@@ -217,19 +217,21 @@ class HuggingFaceCounter(TokenCounter):
 
 
 def _tokenizer_file(repo: str) -> str:
-    """Path to ``repo``'s ``tokenizer.json``, from the hub cache when present.
+    """Path to ``repo``'s ``tokenizer.json``, from the embedding model cache when present.
 
-    A warmed cache (the Docker image bakes the default models) answers without
-    touching the network. ``hf_hub_download`` would otherwise revalidate the
-    revision with a HEAD request on every process start, and stall for the
-    etag timeout on a host that cannot reach huggingface.co.
+    FastEmbed's snapshot of the repository in ``EMBEDDINGS_CACHE_DIR`` already
+    holds the file, so a warmed cache answers without touching the network.
+    ``hf_hub_download`` would otherwise revalidate the revision with a HEAD
+    request on every process start, and stall for the etag timeout on a host
+    that cannot reach huggingface.co.
     """
     from huggingface_hub import hf_hub_download
 
+    cache_dir = settings.EMBEDDINGS_CACHE_DIR or None
     try:
-        return hf_hub_download(repo, "tokenizer.json", local_files_only=True)
+        return hf_hub_download(repo, "tokenizer.json", local_files_only=True, cache_dir=cache_dir)
     except Exception:  # noqa: BLE001 -- not cached: fetch it
-        return hf_hub_download(repo, "tokenizer.json")
+        return hf_hub_download(repo, "tokenizer.json", cache_dir=cache_dir)
 
 
 def _load_hf_counter(repo: str) -> Optional[HuggingFaceCounter]: