feat: add SQLite support behind a sqlite Cargo feature

Adds a SqliteDialect alongside PostgresDialect, gated by mutually exclusive
postgres (default) / sqlite Cargo features, plus a feature-selected
ActiveDatabase type alias used throughout the public Georm/Defaultable
traits. Basic CRUD and Defaultable are fully covered by a new SQLite test
suite (tests/sqlite/*) running against a real SQLite database, alongside
split migrations/sqlite/*.

Two real dialect differences surfaced beyond placeholder syntax:
- SQLite's INTEGER columns type-infer as i64 in sqlx's compile-time query
  macros (unlike Postgres SERIAL -> i32), so the SQLite test models use i64
  for integer PK/FK fields.
- sqlx's SQLite query macros need bind arguments bound to a place rather
  than passed as an inline expression (fixed once, centrally, in
  SqlDialect::generate_relation_lookup).

Composite-key entities with chrono::DateTime columns (tests/composite_key.rs)
are not yet ported to SQLite: SQLite has no TIMESTAMPTZ equivalent, so sqlx
infers TEXT date/time columns as String rather than DateTime<Utc> unless the
query explicitly overrides the column type. That needs per-field type
overrides threaded through the generated queries, left as follow-up.

CI now matrixes over both backends; formatting/audit only run once since
they don't vary by backend. examples/postgres/* stays out of
--no-default-features builds by scoping sqlite lint/test just recipes to
`-p georm` (it depends on georm's default postgres feature via Cargo
feature unification, so building it under --features sqlite at the
workspace level double-activates both dialects).

tests/sqlite/models.rs

@@ -0,0 +1,46 @@
+#![cfg(feature = "sqlite")]
+
+use georm::Georm;
+
+#[derive(Debug, Georm, PartialEq, Eq, Default)]
+#[georm(
+    table = "biographies",
+    one_to_one = [{
+        name = "author", remote_id = "biography_id", table = "authors", entity = Author
+    }]
+)]
+pub struct Biography {
+    #[georm(id)]
+    pub id: i64,
+    pub content: String,
+}
+
+#[derive(Debug, Georm, PartialEq, Eq, Default)]
+#[georm(table = "authors")]
+pub struct Author {
+    #[georm(id)]
+    pub id: i64,
+    pub name: String,
+    #[georm(relation = {entity = Biography, table = "biographies", name = "biography", nullable = true})]
+    pub biography_id: Option<i64>,
+}
+
+impl PartialOrd for Author {
+    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+        Some(self.id.cmp(&other.id))
+    }
+}
+
+impl Ord for Author {
+    fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+        self.id.cmp(&other.id)
+    }
+}
+
+// UserRole (composite key, chrono::DateTime<Utc> column) is deliberately not
+// ported here yet: SQLite's compile-time query macros infer TEXT columns as
+// `String`, not `chrono::DateTime<Utc>` — unlike Postgres's TIMESTAMPTZ, which
+// maps directly. Fixing this needs per-column `query_as!` type overrides
+// (`"assigned_at as \"assigned_at: DateTime<Utc>\""`) threaded through the
+// dialect's generated queries, which is follow-up work, not part of this
+// first SQLite slice. See tests/sqlite/composite_key.rs (not yet added).