fix(core): guard print() calls in run_conversation() against OSError

[teknium1]

Summary

Reimplementation of the fix from PR #1358 by @mr-emmett-one (323 commits stale, had a variable deletion bug). Addresses issue #845.

Problem

In headless environments (systemd services, Docker, nohup), stdout can become unavailable mid-session. Raw print() raises OSError: [Errno 5] Input/output error. The critical failure cascade:

1. print() in quiet_mode display raises OSError
2. The except Exception handler catches it
3. The error handler's own print(f"❌ {error_msg}") also raises OSError
4. This propagates out of run_conversation() entirely
5. Cron scheduler marks the job as error — completed work is never delivered

Fix

• _safe_print() — new static method that wraps print() with try/except OSError. Silently drops output when stdout is broken.
• _vprint() now uses _safe_print() — protects all calls through the verbose print path automatically.
• Raw print() calls in run_conversation() hot path converted to _safe_print(): starting conversation, interrupt, budget exhausted, preflight compression, context cache, conversation completed.
• Error handler (the cascading crash point) gets explicit try/except with logger.error() fallback so diagnostic messages aren't silently lost.

Fixes #845