refactor(database)!: decouple SQLs from backend logic and cache stmts

This decouples SQLs from handler logic, makes it easier for auditing and
caching. It also enables the possibility to switch or support multiple
database backends.

blahd/src/database/tests.rs

@@ -0,0 +1,31 @@
+#![expect(clippy::print_stdout, reason = "allowed in tests for debugging")]
+use super::*;
+
+#[test]
+fn init_sql_valid() {
+    let conn = Connection::open_in_memory().unwrap();
+    conn.execute_batch(INIT_SQL).unwrap();
+
+    // Instantiate view to check syntax and availability of `unixepoch()`.
+    // It requires sqlite >= 3.38.0 (2022-02-22) which is not available by default on GitHub CI.
+    let ret = conn
+        .query_row(
+            "SELECT COUNT(*) FROM `valid_user_act_key`",
+            params![],
+            |row| row.get::<_, i64>(0),
+        )
+        .unwrap();
+    assert_eq!(ret, 0);
+}
+
+#[test]
+fn stmt_cache_capacity() {
+    let src = std::fs::read_to_string("src/database.rs").unwrap();
+    let sql_cnt = src.matches("prepare_cached_and_bind!").count();
+    println!("found {sql_cnt} SQLs");
+    assert_ne!(sql_cnt, 0);
+    assert!(
+        sql_cnt <= STMT_CACHE_CAPACITY,
+        "stmt cache capacity {STMT_CACHE_CAPACITY} is too small, found {sql_cnt} SQLs",
+    );
+}