ERR/E0.1: error-set decls + ! / !Named type exprs (parser)

Parser-only first step of the error-handling stream. No sema/codegen.

- token: `kw_error` keyword (`!` reuses existing `.bang`).
- ast: `ErrorSetDecl { name, tag_names }` + `ErrorTypeExpr { name: ?[] }`
  (null = inferred `!`, non-null = `!Named`); wired into Node.Data and
  declName.
- parser: `parseErrorSetDecl` (comma-separated tags, optional trailing
  comma/`;`) dispatched from parseConstBinding; `!` / `!Named` parsed in
  parseTypeExpr; result-list loop enforces error type as trailing-only;
  hasFnBodyAfterArrow skips `.bang` so failable-return fns are recognised.
- print: new focused AST round-trip printer (decls + type exprs); loud
  `error.UnsupportedNode` otherwise. Registered in root.zig.
- sema/lsp: exhaustive switch arms for the two new nodes.
- tests: 11 inline parser unit tests (shapes + 3 round-trip prints + 2
  trailing-position rejections).

zig build, zig build test, and 254/254 examples green.

src/root.zig

@@ -3,6 +3,7 @@ pub const token = @import("token.zig");
 pub const lexer = @import("lexer.zig");
 pub const ast = @import("ast.zig");
 pub const parser = @import("parser.zig");
+pub const print = @import("print.zig");
 pub const types = @import("types.zig");
 pub const target = @import("target.zig");
 pub const builtins = @import("builtins.zig");