[alii]

Named imports are gone. import scarlet/list.{fold} no longer parses. There is one way to import a module, and you name what you use through it:

import scarlet/http
import scarlet/net/socket as sock

fn route(req http.Request) http.Response { .. }
Ok(sock.Data(bytes)) -> ..

Parser. A .{..} after an import is one error that says what to write: Import the module and name it at the use: `import scarlet/list`, then `list.fold` . The parser then skips the list and keeps the import, so there's no cascade after it.

Removed with it:

• ImportDeclaration::items, and the formatter's rendering of it.
• The compiler's per-item binding.
• ReferenceKind::ImportItem, and the alias_of chain that made {X as Y} its own rename class.
• The unused-import rule's item edge.
• The REPL's item tracking.
• The tree-sitter import_items rule.
• The raw frame-slot load: ValueRef::Slot, Load::Slot in scarlet_ir, and FrameSlot.
  • Only item bindings reached it. I checked by making it fail loudly and running the whole suite: the one test that reached it built it by hand.
  • The new VM already refused to load it.
  • resolve_variable now says Place::ThisFunction for a name the function binds, rather than handing out a slot the elaborator never uses.

Migrated:

• Every .scrl in the repo: the stdlib, the examples, and tests/programs.
  • A throwaway rewrite drove the LSP's own find-references, so a local that shadows an imported name was left alone.
  • Types and constructors were rewritten by name.
  • Every rewritten program passes scarlet check.
  • examples/modules.scrl now shows as on a module, since it used to show item aliases.
• The inline programs in the Rust tests: about 75 import lines.

Tests:

• Converted, where the property still holds with qualified names:
  • the constructor-head exhaustiveness and test-ladder cases (now with qualified heads)
  • type-import rollback (now lib.Color)
  • hover and rename through a qualifier
• Deleted, where it only existed because of named imports:
  • an item alias as a rename class
  • swapped constructor aliases
  • a selective import shadowing a prelude name (env_rollback.rs, so there are 43 test executables now, and the project notes say so)
  • "no member" and "is private" errors at an import item. The qualified versions stay.
• New: a parser test that a named import is exactly one error, and that it names the right qualifier with and without as.

docs/visibility.md no longer lists selective import as a place visibility is checked. The changeset is major.

All seven CI gates pass locally. hawk ran for the host target, and nothing here is gated on a cfg. bun run corpus failed at first because of a merged branch's worktree left inside the repo. I removed it, and then corpus passed.