fix(fuzz): fuzz_settings_replay builds again; a B-27 check for an unknown convention kind

The weekly run's fuzz_settings_replay job failed to build since P3 gave
each projection schema its own file: the target still named
ledger_store::DB_FILE, which became db_file(). The fuzz crate is outside
the workspace, so just check never compiled it; every target builds
again (cargo fuzz build, 2026-10-29).

The same run's fuzz_store_replay crash (the writer refusing a journal
with an unknown convention kind as newer while the viewer said something
else) is not reproduced yet: the hand-made case, kept here as
an_unknown_convention_kind_is_newer_for_the_writer_and_the_viewer_alike,
passes, so the divergence needs the fuzzer's exact journal, which the
weekly job now keeps as an artifact.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HS8WZziDnyxunC23MUxrLp

fuzz/fuzz_targets/fuzz_settings_replay.rs

@@ -167,7 +167,7 @@ fuzz_target!(|data: &[u8]| {
             w.append(t, p).unwrap();
         }
     }
-    let _ = std::fs::remove_file(dir.join(ledger_store::DB_FILE));
+    let _ = std::fs::remove_file(dir.join(ledger_store::db_file()));
     let store = ledger_store::Store::open(&dir);
     let snap = ledger_store::Snapshot::load(&dir);
     match (store, snap) {