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

[teyrebaz33]

Closes #845

Problem

When hermes-agent runs as a systemd service or headless daemon, the stdout pipe can become unavailable (idle timeout, buffer exhaustion, socket reset), causing print() to raise OSError: [Errno 5] Input/output error. This crashes run_conversation() and causes cron jobs to be marked as failed.

Changes

Wrap 7 affected print() calls in try/except OSError in run_agent.py:

Line	Description	Fallback
~3664	Context length discovery	pass
~3694	Interrupt during API call	pass
~3737	API retry warning	logger.warning
~3922	API failed after N attempts	logger.warning
~3958	All retries exhausted	logger.error
~4191	quiet_mode 💬 display	pass
~4360	Error handler ❌ message	logger.error

Cosmetic lines are silently dropped. Error handler lines fall back to logger so the message is preserved.

Testing

3 new tests in TestPrintOSErrorGuard. Full suite: 2863 passed.

[teyrebaz33]

The failing check (test_vision_tools.py::TestErrorLoggingExcInfo::test_analysis_error_logs_exc_info) is unrelated to this PR — fix is in #804.

[teknium1]

Closing — the approach of wrapping individual print() calls doesn't scale. There are 68 print() calls in run_conversation() alone, and this PR guards only 7 of them (leaving adjacent prints in the same blocks unguarded). We're implementing a systematic fix that covers all of them without touching each call site. Thanks for bringing this to our attention — the issue is real and the root cause analysis in #845 was excellent.