fix(io): use a context manager in _lesson_already_appended

[diazMelgarejo]

Atomic PR 3 of 3 (PR 1, episodic-mirror, merged as #57; PR 2, UTF-8 fixes, #61 — this PR is built on top of it, so its diff includes #61's commit until that one merges).

Summary

_lesson_already_appended() opened lessons_path with a bare open() and never closed it explicitly — relying on GC/refcounting to close the file handle. Same resource-safety class as the file-handle findings from this project's own recent hygiene passes. Wraps the read in a with open(lessons_path, encoding="utf-8") as f: block; also picks up explicit UTF-8 encoding, consistent with #61.

Tests

Same suite as #61 (built on top of it): 177 passed, 1 pre-existing unrelated failure (test_unknown_executable_is_a_structured_start_failure).

Please merge #61 first — once it lands, this PR's diff collapses to just this fix's own delta.

Note

Fix file handling in _lesson_already_appended to use a context manager with UTF-8 encoding

Updates learn.py to consistently use UTF-8 encoding across file reads and writes, preventing locale-dependent encoding errors.

• _lesson_already_appended now opens lessons.jsonl with a context manager and explicit UTF-8 encoding instead of relying on system defaults.
• stage writes candidate JSON files with explicit UTF-8 encoding to preserve non-ASCII content.
• sys.stdout and sys.stderr are reconfigured to UTF-8 with errors='replace' at module load time.
• Behavioral Change: files previously read or written using the system locale encoding will now always use UTF-8; this may surface previously hidden encoding mismatches.

^{Macroscope summarized f685a08.}

[diazMelgarejo]

Related, independent PRs from the same review pass — each stands on its own, no dependency between them:

• fix(recall): exclude superseded lessons from lessons.jsonl retrieval #60 — recall.py excludes superseded lessons from lessons.jsonl retrieval
• fix(io): force UTF-8 on stdout/stderr and candidate writes #61 — UTF-8 stdout/stderr + candidate writes in learn.py (this PR is built on top of fix(io): force UTF-8 on stdout/stderr and candidate writes #61's commit on the fork side, so its diff will collapse to just its own delta once fix(io): force UTF-8 on stdout/stderr and candidate writes #61 merges — worth landing fix(io): force UTF-8 on stdout/stderr and candidate writes #61 first)
• fix(upgrade): loop-* skill copy doubles the skills/ path segment #63 — unrelated bug found while reviewing this same area: harness_manager/upgrade.py's loop-skill installer doubles a path segment

cc @codejunkie99 — happy to answer questions on any of these, or squash/reorder however's easiest to review.