Deploy: LanguageTool spell checker in production, with optional n-grams

- docker-compose.yml: a `languagetool` service (erikvl87/languagetool:6.8),
  internal only, 1 GB heap, one CPU, a 28 GB memory limit that is room for the
  n-gram page cache (reclaimable, not memory in use), and a low-priority
  warm-up read of the n-gram files at start so first checks hit RAM even on
  mechanical disks. `init` creates $DATA_ROOT/languagetool-ngrams; the app
  containers get LANGUAGETOOL_URL (empty turns the check off).
- `webmail.sh ngrams [langs]` (default en de fr es): deploy/ngrams.py, run in
  the backend image, downloads LanguageTool's n-gram data with resume, checks
  free space first (unpacked is ~2x the zip: ~26 GB for all four), unpacks into
  a temp folder and moves it into place, then recreates the engine to load it.
  languagetool.org refuses Python's default User-Agent, so it sends its own.
- `doctor` reports whether the spell checker answers and which n-gram
  languages are installed.
- docs/nas-install.md and .env.prod.example describe all of it.

Verified locally: German n-grams (1.58 GB -> 3.06 GB) load, and catch what the
rules alone do not ("liegt ihm Schrank" -> im, "Der Patenten" -> Patienten).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/nas-install.md

@@ -606,6 +606,38 @@ Its Bayes database lives in `$DATA_ROOT/rspamd-redis`. Like the search index,
 that is derived data: it rides along in backups but never needs restoring, and
 retraining rebuilds it.
 
+### The spell checker (LanguageTool)
+
+`languagetool` checks spelling and grammar in the compose window, in some 30
+languages. Like the anti-spam engine nothing waits for it: when it is down the
+compose window falls back to the browser's own spell checker. It is never
+published to the host — the app's API is its only client.
+
+It needs about **1.2 GB of RAM** (`LANGUAGETOOL_HEAP=1g` plus the JVM) and is
+capped at one CPU core (`LANGUAGETOOL_CPUS`), so checking can never slow sync,
+search or the API.
+
+**Optional n-gram data** teaches it to catch real words used wrongly — "the
+affect was huge", "Das Buch liegt ihm Schrank" — beyond its hand-written rules
+(English has ~780 such word pairs, German ~70). Install it once:
+
+```bash
+./deploy/webmail.sh ngrams            # en de fr es — or name languages: ngrams en de
+```
+
+That downloads about **13 GB** (English 8.4 GB, German/French/Spanish ~1.6 GB
+each) into `$DATA_ROOT/languagetool-ngrams`, unpacks it to about **26 GB**, and
+restarts the engine to load it. An interrupted download resumes when you run
+the command again. There is no n-gram data for Polish; Polish text is checked
+exactly as without it.
+
+The data is read from disk on demand and cached by the kernel, so the memory
+limit (`LANGUAGETOOL_MEM_LIMIT=28g`) is room for that cache, not memory in use.
+At start the container reads the files once in the background, at low
+priority, so even the first checks after a restart are served from RAM. On
+mechanical disks that warm-up takes several minutes; an NVMe cache in front of
+the volume makes it — and any later cache misses — fast.
+
 Two NAS-specific realities worth knowing:
 
 - **Disk hibernation won't happen.** Active accounts hold IMAP IDLE connections
@@ -839,6 +871,7 @@ directories are root-owned.
 ./deploy/webmail.sh backup           verified dump, right now
 ./deploy/webmail.sh restore <dump>   restore a dump (asks for confirmation)
 ./deploy/webmail.sh reindex [--purge] rebuild the search index
+./deploy/webmail.sh ngrams [langs]   spell-check n-gram data (default: en de fr es)
 ./deploy/webmail.sh psql             database shell
 ./deploy/webmail.sh shell [svc]      shell in a container
 ```