backup: #904 v20 - fix TestEncodeSnapshotRejectsZeroAdapterSet to pin the right guard

Claude v19 caught: the v19 MANIFEST.json gate (added to checkInputRoot)
moved this test's failure point from the zero-adapter guard (the
invariant the test name and comment claim it pins) to the
MANIFEST.json-missing guard. The test still passed both assertions
(non-nil error, buf.Len()==0) but for the wrong reason.

Adding AllowMissingManifest: true so checkInputRoot waves through the
missing manifest and validateEncodeOptions proceeds to the
zero-adapter check that's the actual subject of the test.

The other 10 synthetic-fixture EncodeSnapshot callers in v19 already
got this flag; this one was missed because its rejection assertions
were satisfied by the new guard, so the failure wasn't a hard error.

Test-only; no behavior change. Tests + lint green.

Author: bootjp

internal/backup/encode_snapshot_test.go

@@ -436,6 +436,13 @@ func TestEncodeSnapshotRejectsZeroAdapterSet(t *testing.T) {
 		InputRoot:    in,
 		Adapters:     AdapterSet{}, // explicit zero
 		LastCommitTS: 1,
+		// AllowMissingManifest: true bypasses the v19 MANIFEST.json
+		// guard so this test actually exercises the zero-adapter
+		// guard further down in validateEncodeOptions — without this,
+		// checkInputRoot would error on the missing MANIFEST.json
+		// first and the assertion would pin the wrong invariant
+		// (claude v19 #904).
+		AllowMissingManifest: true,
 	}, &buf)
 	if err == nil {
 		t.Fatalf("EncodeSnapshot with empty AdapterSet succeeded; want error")