Bump version to 0.4.1; pin upeepz80>=0.2.3

Bump version to 0.4.0

~20 commits of real work since 0.3.0:

- Tasking: scheduler reentrancy, halt-when-no-tasks-ready, entry
  call/accept rendezvous + parameter passing, IX clobbering on accept,
  select/terminate alternative, DECLARE-block task vars, conditional
  entry calls, selective accept entry matching, delay alternatives,
  Duration type, zero-tick delay yields instead of blocking forever,
  implicit task exception handlers, guard conditions, task attribute
  prefix fix, E'Count, all-closed PROGRAM_ERROR, select_entry_cnt
  2-byte fix.
- Generic subtype lowering and STANDARD.X function calls.
- Protected types in packages, rendezvous semantics, task attributes,
  record field / protected type resolution in lowering.
- String ordering, record aggregate init, discriminant-dependent
  sizing, outer array bounds, aggregate comparison, qualified expr,
  SubtypeDecl, adalib parse errors / support file resolution / array
  bounds safety.
- Docs: drop stale 8080 target claims (compiler is Z80-only).

Also sync uada80/__init__.py __version__ (was lagging at 0.2.0).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>