ffi 2.7 green: parser accepts all seven type-introducer directive forms

Six new lexer tokens (`hash_jni_interface`, `hash_objc_class`,
`hash_objc_protocol`, `hash_swift_class`, `hash_swift_struct`,
`hash_swift_protocol`) join the existing `hash_jni_class`. All seven
share the body grammar from Phases 2.1–2.6.

AST refactored: `JniClassDecl` → `ForeignClassDecl` with a
`runtime: ForeignRuntime` enum discriminator; `JniMethodDecl` →
`ForeignMethodDecl` (with `jni_descriptor_override` renamed for
clarity since it's JNI-only); `JniFieldDecl` → `ForeignFieldDecl`;
`JniClassMember` → `ForeignClassMember`. AST variant renamed
`jni_class_decl` → `foreign_class_decl`.

`parseForeignClassDecl` takes the runtime as a parameter; the
`parseConstBinding` dispatch table now maps each of the seven
directive tokens to its `ForeignRuntime` variant via
`foreignRuntimeForCurrent`. No codegen yet — Phase 3 picks up Obj-C
runtime, Phase 4 picks up Swift. Runtime-specific body items (fields,
descriptor override) are validated at sema time in later steps.

126/126 examples green.

src/sema.zig

@@ -907,8 +907,8 @@ pub const Analyzer = struct {
                     }
                 }
             },
-            .jni_class_decl => |jd| {
-                try self.addSymbol(jd.name, .type_alias, null, node.span);
+            .foreign_class_decl => |fd| {
+                try self.addSymbol(fd.name, .type_alias, null, node.span);
             },
             .impl_block => |ib| {
                 // Each impl block gets its own scope so methods don't conflict across impls
@@ -1309,7 +1309,7 @@ pub fn findNodeAtOffset(node: *Node, offset: u32) ?*Node {
         .tuple_type_expr,
         .ufcs_alias,
         .closure_type_expr,
-        .jni_class_decl,
+        .foreign_class_decl,
         => {},
         .struct_decl => |sd| {
             for (sd.methods) |method_node| {