fix(media-verify): read the WAL the server commits through

`mode=ro&immutable=1` claims the file never changes, so sqlite skips the
write-ahead log entirely: every row the server has committed but not yet
checkpointed reads as MISSING, and a read beside a running scanner can fail
outright as "database disk image is malformed". Both servers keep their
libraries in WAL, so plain `mode=ro` is now the first door and `immutable=1`
only the fallback for a database whose `-shm` cannot be read.

The verdict also stopped guessing from the database alone. A file still at its
`src` is REAL (this pass's own damage); a file at neither path is `gone` (a
later pass or the server renamed it); a file sitting at `dst` with no row
naming it is something the server does not index. A row that still names `src`
is none of those — it is the server's stale pointer, and it costs a re-scan,
never a restore.

scripts/media-verify.py

@@ -9,8 +9,13 @@ its destination, and NO row may still name its source — a source row is a scan
 that has not caught up, and it is the one signature of a library pointing at
 files that no longer exist.
 
-Two doors, both read-only, both without a lock while the server runs
-(`mode=ro&immutable=1`; there is no `sqlite3` binary on these hosts):
+Two doors, both read-only, both without a lock while the server runs (there is
+no `sqlite3` binary on these hosts). Open `mode=ro` PLAIN: both servers keep
+their libraries in WAL mode, and `immutable=1` — which claims the file never
+changes — makes sqlite skip the write-ahead log entirely, so every row the
+server has committed but not yet checkpointed reads as MISSING, and a read
+beside a running scanner can even fail as "database disk image is malformed".
+`immutable=1` is only the fallback, for a database whose `-shm` cannot be read.
 
   jellyfin  `BaseItems.Path`. The container binds the share at `/data`, so the
             database holds `/data/<library>/…` where the journal holds
@@ -22,9 +27,14 @@ Two doors, both read-only, both without a lock while the server runs
 
 Neither server holds a row for everything the journal renames: a `.log`, `.cue`
 or cover image has no row to land in, so its absence is not a fault. The report
-therefore breaks the missing rows down by extension, and `--ignore-ext` names
-the extensions this server does not index, which are then counted separately and
-kept out of the verdict.
+therefore breaks the missing rows down by extension, and `--ignore-ext` names the
+extensions this server does not index, which are then counted separately and kept
+out of the verdict. What is left is split by what the filesystem says, because
+the database alone cannot: a file still at its `src` is `REAL` (this pass's own
+damage), a file at neither path is `gone` (a later pass or the server renamed it),
+and a file sitting at its `dst` with no row naming it is simply something the
+server does not index. A row that still names `src` is none of these: it is the
+server's own stale pointer — a phantom — and it costs a re-scan, never a restore.
 
 Exit codes: 0 every move landed; 1 something is stale or missing (the details are
 printed); 2 the journal or the database is not usable at all.
@@ -92,10 +102,18 @@ def read_moves(path):
 def server_paths(db, kind):
     """Every path the server holds, or a refusal that names what is missing."""
     spec = KINDS[kind]
-    try:
-        con = sqlite3.connect(f"file:{db}?mode=ro&immutable=1", uri=True)
-    except sqlite3.Error as e:
-        die(f"cannot open {db}: {e}")
+    con = None
+    err = None
+    # Plain ro reads the WAL the server commits through; immutable is the last
+    # resort, because it cannot see a row the server has not checkpointed.
+    for opt in ("mode=ro", "mode=ro&immutable=1"):
+        try:
+            con = sqlite3.connect(f"file:{db}?{opt}", uri=True)
+            break
+        except sqlite3.Error as e:
+            err = e
+    if con is None:
+        die(f"cannot open {db}: {err}")
     with con:
         have = {r[1] for r in con.execute(f"pragma table_info({spec['table']})")}
         if not spec["columns"] <= have:
@@ -161,20 +179,37 @@ def main():
         if at_src:
             stale.append(src)
         if not at_dst:
-            missing.append(dst)
+            missing.append((src, dst))
+
+    # The database cannot tell a lost file from one it never indexes: both read
+    # as "no row at the destination". Stat both paths, so the report separates
+    # the classes the runbook names — `REAL` is the only one that is this pass's
+    # own damage, and `gone` is a later pass or the server's own rename.
+    on_disk, real, gone = [], [], []
+    for src, dst in missing:
+        if os.path.exists(dst):
+            on_disk.append(dst)
+        elif os.path.exists(src):
+            real.append(src)
+        else:
+            gone.append(dst)
 
     print(f"{args.db_kind}: {len(moves)} moves checked")
     print(f"landed: {landed}")
     print(f"still at src: {len(stale)}")
     print(f"missing from the DB: {len(missing)}")
+    print(f"  on disk at dst, no row names it: {len(on_disk)}")
+    print(f"  REAL — file still at src: {len(real)}")
+    print(f"  gone — at neither path: {len(gone)}")
     if ignored:
         print(f"not indexed (extension ignored): {ignored}")
     if missing:
         exts = Counter(os.path.splitext(d)[1].lower() or "(no extension)"
-                       for d in missing)
+                       for _, d in missing)
         print("missing by extension: "
               + ", ".join(f"{e} {n}" for e, n in sorted(exts.items())))
-    for label, items in (("stale", stale), ("missing", missing)):
+    for label, items in (("real", real), ("gone", gone), ("missing", on_disk),
+                         ("stale", stale)):
         for path in items[:args.show]:
             print(f"  {label}: {path}")
         if len(items) > args.show: