ffi 2.1: parser accepts Foo :: #jni_class("path") { } opaque form

New `hash_jni_class` token + lexer entry, `JniClassDecl` AST node
(alias + java path; body deferred to 2.2+), `parseJniClassDecl`
consuming `("...") { }` and rejecting non-empty bodies for now.
Sema registers the alias as a type_alias symbol; LSP classifies
the directive as a keyword. The 2.0 xfail snapshot flips to
`parse-only ok`, exit 0.

120/120 examples green; zig test clean.

src/sema.zig

@@ -907,6 +907,9 @@ pub const Analyzer = struct {
                     }
                 }
             },
+            .jni_class_decl => |jd| {
+                try self.addSymbol(jd.name, .type_alias, null, node.span);
+            },
             .impl_block => |ib| {
                 // Each impl block gets its own scope so methods don't conflict across impls
                 try self.pushScope();
@@ -1306,6 +1309,7 @@ pub fn findNodeAtOffset(node: *Node, offset: u32) ?*Node {
         .tuple_type_expr,
         .ufcs_alias,
         .closure_type_expr,
+        .jni_class_decl,
         => {},
         .struct_decl => |sd| {
             for (sd.methods) |method_node| {