fix: add explicit UTF-8 encoding to read_text() calls (#776)

On Windows with non-UTF-8 locale (e.g. GBK), Path.read_text() defaults
to platform encoding, breaking onboarding tests and any source code that
reads JSON/markdown with non-ASCII content.

5 files, 8 call sites fixed.

tests/test_instructions_cli.py

@@ -10,7 +10,7 @@ from mempalace.instructions_cli import AVAILABLE, INSTRUCTIONS_DIR, run_instruct
 def test_run_instructions_valid_name(capsys):
     """Valid name prints the .md file content."""
     name = "init"
-    expected = (INSTRUCTIONS_DIR / f"{name}.md").read_text()
+    expected = (INSTRUCTIONS_DIR / f"{name}.md").read_text(encoding="utf-8")
     run_instructions(name)
     captured = capsys.readouterr()
     assert captured.out.strip() == expected.strip()