Improve large-run resilience: checkpoint timeouts + CSV save fallback

[hemanth-asirvatham]

Motivation

• Long runs can hit timeouts and produce huge final CSVs that fail to save; make resumed runs reuse prior timing info and avoid crashing on writes.
• The previous large-run warning at 50k rows was too conservative and the message could be friendlier.
• Provide a simple, reusable fallback for writing very large DataFrames so tasks return results even when disk saves fail.

Description

• Raise the large-run advisory threshold in get_all_responses from 50k to 100k and soften the message to suggest splitting only if API/timeout issues are encountered, using get_all_responses parameters for context.
• Add checkpoint-aware timeout bootstrapping by extracting successful Time Taken values from existing checkpoint CSVs and initializing the dynamic timeout via new helpers _collect_successful_time_taken_samples and _compute_dynamic_timeout_from_samples in src/gabriel/utils/openai_utils.py.
• Introduce save_dataframe_with_fallback in src/gabriel/utils/file_utils.py (exported via gabriel.utils) which attempts a normal DataFrame.to_csv and on failure writes 100k-row split files (<stem>_1.csv, <stem>_2.csv, ...), logging and printing informative messages without raising.
• Wire the fallback saver into task outputs that commonly emit final/aggregated CSVs (rate, classify, extract) and into other final-output tasks where simple to do (deidentify, paraphrase, codify, seed, debias) to improve robustness on large outputs.

Testing

• Added unit tests for the timeout helpers (tests/test_timeout_logic.py) and for checkpoint resume behaviour (tests/test_checkpoint_timeout_resume.py) which validate that resumed runs can bootstrap dynamic timeouts from checkpoint Time Taken samples.
• Added tests for CSV fallback behaviour (tests/test_save_dataframe_fallback.py) covering successful split fallback and all-writes-fail scenarios.
• Ran full test suite and compilation: python -m compileall -q src tests and pytest -q completed successfully with all tests passing (188 passed, 6 skipped).

---

Codex Task

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}