BlockingIOError on large agent prompt logging in non-interactive / piped environments

[hertznsk]

Describe the bug

When running a workflow in a non-interactive environment (such as cron or when stderr/stdout is connected to a pipe), conductor run fails with BlockingIOError: [Errno 11] write could not complete without blocking if an agent's rendered prompt is large (e.g. > 64 KB).

Context & Reproduction Details

• Workflow scenario: A multi-agent workflow where parallel collector agents gather data and pass aggregated outputs to a downstream synthesis agent.
• Prompt size: The rendered prompt for the synthesis agent was 58,161 characters (~105 KB UTF-8).
• Execution environment: Running via cron (0 7 * * * /bin/bash -lc "... conductor run workflow.yaml --no-interactive").
• In cron/piped environments, standard streams are pipes without a TTY. When writing large continuous blocks exceeding the OS pipe buffer capacity (typically 64 KB on Linux) to a pipe descriptor in non-blocking mode, sys.stderr.write() / self.file.write() raises BlockingIOError.

Observed Behavior & Stack Trace

1. Orchestrator completes collector agents, routes to synthesis, prepares the agent prompt, and attempts to log it via verbose_log_section.
2. Logging to _verbose_console / Rich Console fails on self.file.write(text):

  File ".../site-packages/conductor/executor/agent.py", line 490, in execute
    _verbose_log_section(
        f"Prompt for '{agent.name}'",
        rendered_prompt,
    )
  File ".../site-packages/conductor/executor/agent.py", line 40, in _verbose_log_section
    verbose_log_section(title, content)
  File ".../site-packages/conductor/cli/run.py", line 401, in verbose_log_section
    _verbose_console.print(
        Panel(
            Text(content),
            title=styled("[cyan]{}[/cyan]", title),
            border_style="dim",
        )
    )
  ...
  File ".../site-packages/rich/console.py", line 2124, in _write_buffer
    self.file.write(text)
BlockingIOError: [Errno 11] write could not complete without blocking

3. The engine catches the exception and attempts error recovery / saving checkpoint, but the teardown handlers (_print_resume_instructions and print_error) also invoke console printing, triggering repeated BlockingIOError exceptions.

Suspected Root Cause (Presumed)

• verbose_log_section unconditionally prints the full rendered prompt Panel to console when in full verbose mode (which is default).
• Diagnostic / verbose logging to standard output/error streams does not handle BlockingIOError / non-blocking pipe semantics.
• A failure during verbose logging aborts the entire workflow execution rather than being handled gracefully or suppressed.

Expected Behavior

• Verbose console logging should not crash the workflow execution if write() cannot complete immediately or if the pipe buffer is full.
• Error recovery handlers (_print_resume_instructions, print_error) should survive stream write errors without crashing.