feat: add unified context-aware logging system

[devin-ai-integration]

Summary

Replaces all inconsistent logging approaches (13+ raw print() calls, manual [DEBUG]/[INFO] tags, sys.stderr writes, and the partial log_callback: Callable abstraction) with a unified, context-aware, hierarchical logging system.

New package src/crab/log/:

• logger.py — CrabLogger with hierarchical context nesting via enter(), thread-safe emit with a shared lock, live subprocess stdout streaming via background threads
• formatters.py — RichFormatter (ANSI-colored tree-style) and PlainFormatter (grep-friendly brackets)
• handlers.py — StreamHandler (stdout → slurm_output.log under SLURM), TUIHandler (Rich markup → Textual RichLog widget)
• __init__.py — get_logger() factory, reads CRAB_LOG_LEVEL env var

Integration changes:

• engine.py — Engine and ExperimentRunner accept CrabLogger instead of Callable. run_job() and end_job() accept a logger. Live streaming enabled for concurrent apps via stream_process() background threads.
• orchestrator.py — All print() replaced. New --log-level CLI flag added. Workers read CRAB_LOG_LEVEL from environment.
• controller.py — TUIController builds a CrabLogger with a TUIHandler wired to the RichLog widget callback, replacing raw Rich markup strings in log calls.
• slurm.py / mpi.py — Debug prints removed; wl_managers stay logger-agnostic.
• Wrappers (microbench_common.py, miniFE.py, ib_send_lat.py) — Stray print() calls removed.

Review & Testing Checklist for Human

• Signature compatibility at all call sites: run_job(), end_job(), wait_timed(), Engine(), and ExperimentRunner() all have changed signatures. Verify no call site was missed (search for old patterns like log_callback=, log_fn=, run_job( without logger=). A missed call site will cause a runtime crash on the cluster.
• Dynamic _stream_thread attribute on job objects: job._stream_thread is set dynamically (not declared on the job/app class). Verify this doesn't conflict with existing attributes and that all hasattr guards are sufficient. Consider whether this should be a proper attribute on the base class.
• Concurrent app detection logic: concurrent = len(static_schedule) > 1 or len(dependency_map) > 0 determines whether to live-stream stdout. Verify this correctly identifies concurrent vs sequential scenarios for your experiment configurations. A false positive means unnecessary threads; a false negative means no live output for concurrent apps.
• TUI dual-handler behavior: TUIController now has both a StreamHandler (from get_logger()) AND a TUIHandler. This means TUI mode logs to both stdout and the widget. Verify this is acceptable or if the StreamHandler should be omitted in TUI mode.
• End-to-end test on SLURM: Submit a real benchmark job and verify (a) slurm_output.log contains properly formatted, colored output, (b) tail -f slurm_output.log renders colors correctly, (c) concurrent app output is interleaved correctly with context labels, (d) --log-level DEBUG shows debug output, and (e) the TUI log tab displays records correctly.

Notes

• traceback.print_exc() calls in fatal error handlers still write directly to stderr — this is intentional since the logger itself may be in a broken state at that point.
• The workerpool_scheduler.py logging was intentionally left untouched per requirements (it's a separate sub-benchmark with its own scheduler.log).
• The BenchmarkState import was removed from controller.py — verify it was truly unused before merging.

Link to Devin session: https://app.devin.ai/sessions/fc7b895897f54186a4d154c93d00b658
Requested by: @SharkGamerZ

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring