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

[mvalentsev]

What

Path.read_text() without encoding defaults to platform encoding on Windows. On locales like GBK this breaks any file written as UTF-8.

8 call sites fixed across 5 files:

• tests/test_onboarding.py (4x) -- the reported bug: onboarding writes UTF-8 (lines 315, 362), tests read without encoding
• tests/test_instructions_cli.py (1x) -- reads UTF-8 markdown
• mempalace/entity_registry.py (1x) -- reads JSON with non-ASCII entity names
• mempalace/split_mega_files.py (1x) -- reads JSON with non-ASCII known names
• mempalace/instructions_cli.py (1x) -- prints UTF-8 markdown to terminal

Why explicit encoding, not PYTHONUTF8

Considered alternatives:

• PYTHONUTF8=1 in CI -- must be set before Python starts, so it can't go in conftest.py. Works for CI but doesn't fix local test runs on Windows. Also a global override of all I/O which can mask real encoding bugs.
• PEP 686 (Python 3.15) -- UTF-8 will become the default, but mempalace supports >=3.9 so that's years away.
• Explicit encoding="utf-8" per call -- PEP 597 recommendation. Works on all Python versions, all platforms, no env setup needed.

Test plan

• CI passes on all platforms (especially test-windows)
• pytest tests/test_onboarding.py tests/test_instructions_cli.py -v -- 52 passed locally