fix(workspace): write and read the vault as UTF-8, not the host locale

[MohammedAlkindi]

AntonSettings loads .anton/.env with env_file_encoding="utf-8" (anton/config/settings.py:32), but Workspace wrote and read that file, and anton.md, with bare read_text / write_text. Those use locale.getpreferredencoding(), which is cp1252 on a Western Windows install, so the writer and the loader disagree.

Two consequences, both silent until they are not:

• a non-ASCII secret is stored as bytes the loader cannot decode, so every later AntonSettings() raises UnicodeDecodeError;
• a value outside the code page raises UnicodeEncodeError on write and is never stored at all.

Five tests cover it. Four are behavioural; the fifth runs the workspace under -X warn_default_encoding (PEP 597) and asserts no EncodingWarning names workspace.py, which keeps the regression visible on UTF-8 CI where the behavioural four pass either way.

All five fail on the unpatched tree and pass with the change.

Scoped to the nine test modules that reach Workspace or cli: 403 passed / 9 failed before, 408 passed / 9 failed after, with a byte-identical set of failing names. The nine are pre-existing clipboard, network and POSIX-permission tests unrelated to this change.

The same two lines applied to the consent write in cli._ensure_terms_consent.