Merge 'perf/throughput: force sqlite to use fullfsync' from Pedro Muniz

I was sampling our performance and noticed that In the throughput test,
we were only setting the `PRAGMA synchronous = full` in `setup_database`
and not in the connection. Also we were not setting `PRAGMA fullfsync =
true`.
Now SQLite numbers and Turso numbers are much closer:
`Turso,1,1000,0,215095.94`
vs
`SQLite,1,1000,0,219748.39`
Specs: M2 Air 8gb
related: #3027

Closes #3734

perf/throughput/rusqlite/src/main.rs

@@ -94,6 +94,7 @@ fn setup_database(db_path: &str) -> Result<Connection> {
 
     conn.pragma_update(None, "journal_mode", "WAL")?;
     conn.pragma_update(None, "synchronous", "FULL")?;
+    conn.pragma_update(None, "fullfsync", "true")?;
 
     conn.execute(
         "CREATE TABLE IF NOT EXISTS test_table (
@@ -116,6 +117,9 @@ fn worker_thread(
 ) -> Result<u64> {
     let conn = Connection::open(&db_path)?;
 
+    conn.pragma_update(None, "synchronous", "FULL")?;
+    conn.pragma_update(None, "fullfsync", "true")?;
+
     conn.busy_timeout(std::time::Duration::from_secs(30))?;
 
     start_barrier.wait();