Add the subtitle schema

Vellum had no subtitle support of any kind. This is the storage half.

Three sources, and they are genuinely different shapes: a sidecar file beside
the video has a path, an embedded stream has an index into the container and
no file of its own, and a downloaded track has a path but is regenerable.

stream_index is -1 for the file-backed sources rather than NULL, and that is
deliberate: NULL is never equal to NULL in SQL, so a unique index over a
nullable column would happily accept the same sidecar twice. With -1 the
UNIQUE (item_id, source, path, stream_index) actually does its job and a
rescan can upsert.

Nothing here is authoritative about the media. Deleting every row and
rescanning reproduces it exactly, which is what lets a vanished subtitle file
have its row removed - unlike a vanished video, which is marked instead
(ADR-003) because progress hangs off it and cannot be regenerated.

Verified on both paths: a real database at version 13 upgraded to 14 by an
older binary-then-newer sequence, and a fresh database applied 0001..0014.

migrations/0014_subtitles.sql

@@ -0,0 +1,45 @@
+-- 0014_subtitles.sql — subtitle tracks per item.
+--
+-- Vellum had no subtitle support at all. This is the storage half: a row per
+-- track, whether it came from a file next to the video, a stream muxed into it,
+-- or a download.
+--
+-- Three sources, and they are genuinely different shapes:
+--
+--   * 'sidecar'    a file beside the video (".en.srt"). Has a path, no stream.
+--   * 'embedded'   a stream inside the container. Has a stream index, no path
+--                    of its own — it only exists inside the video file.
+--   * 'downloaded' fetched from a provider, cached under the config dir. Has a
+--                    path, like a sidecar, but is regenerable and can be
+--                    deleted without losing anything.
+--
+-- `stream_index` is -1 for the two file-backed sources rather than NULL, so the
+-- UNIQUE below can do its job: NULL is never equal to NULL in SQL, and a unique
+-- index over a nullable column would happily accept the same sidecar twice.
+--
+-- `language` is BCP-47 where we can tell ("en", "pt-BR") and empty when we
+-- cannot. It is a display and preference key, not a correctness one: a track
+-- with no language still plays, it just cannot be auto-selected.
+--
+-- This table is a cache of what is on disk. Deleting every row and rescanning
+-- must reproduce it exactly, which is why nothing here is authoritative about
+-- the media itself.
+
+CREATE TABLE subtitles (
+    id           INTEGER PRIMARY KEY,
+    item_id      INTEGER NOT NULL REFERENCES media_items(id) ON DELETE CASCADE,
+    source       TEXT    NOT NULL,                -- 'sidecar' | 'embedded' | 'downloaded'
+    path         TEXT    NOT NULL DEFAULT '',     -- relative to the library root; empty when embedded
+    stream_index INTEGER NOT NULL DEFAULT -1,     -- ffmpeg stream index; -1 when file-backed
+    format       TEXT    NOT NULL DEFAULT '',     -- 'srt' | 'ass' | 'ssa' | 'vtt' | 'sub'
+    language     TEXT    NOT NULL DEFAULT '',     -- BCP-47 where known
+    label        TEXT    NOT NULL DEFAULT '',     -- what the client shows; the filename or stream title
+    bytes        INTEGER NOT NULL DEFAULT 0,
+    added_at     TEXT    NOT NULL DEFAULT (strftime('%Y-%m-%dT%H:%M:%fZ','now')),
+
+    UNIQUE (item_id, source, path, stream_index)
+);
+
+-- The read is always "every track for this item", and it happens on every
+-- player open, so it is the one index worth having.
+CREATE INDEX idx_subtitles_item ON subtitles(item_id, source, language);