fix(qlib): stop dropping read_exp_res.py's log on failed backtest runs

[Pujitha Paladugu (pujitha24)]

Description

QlibFBWorkspace.execute() runs qrun <config> (captured as execute_qlib_log) and then python read_exp_res.py (captured as execute_log) inside the target docker/conda environment. read_exp_res.py is what raises LoadObjectError: No such file or directory: '.../report_normal_1day.pkl' when an expected backtest artifact is missing.

When the resulting ret.parquet or qlib_res.csv file doesn't exist, execute() returned only execute_qlib_log, silently dropping execute_log — the log that actually contains read_exp_res.py's failure reason. This return value is surfaced verbatim to the user/LLM feedback loop via raise FactorEmptyError(f"Failed to run this experiment, because {stdout}") in factor_runner.py (and the analogous path in model_runner.py), so the real cause of the failure was thrown away.

This change concatenates execute_log into both failure-path return values so read_exp_res.py's own output is preserved alongside the qrun log.

This does not fix the underlying reason an artifact may go missing (e.g. an environment-specific GPU/CUDA problem, as diagnosed in a comment on the linked issue) — it only preserves the diagnostic information needed to explain that, instead of discarding it.

Motivation and Context

Report: #442

The reporter saw the workflow stall with a confusing downstream error (LoadObjectError on report_normal_1day.pkl) and no visible root cause. A commenter traced their own case to a CUDA/GPU environment issue causing the backtest artifact to never be generated — but regardless of root cause, RD-Agent's own error-reporting path was silently discarding the one log (read_exp_res.py's output) that would show why the artifact is missing, making this class of failure needlessly hard to diagnose.

How Has This Been Tested?

• Added test/qlib/test_workspace.py, which mocks the docker/conda environment and asserts that when the backtest result file is missing, the returned log contains both the qrun log and read_exp_res.py's own failure output (in this repo's existing unittest.mock-based pytest style, matching test/qlib/test_model_factor_proposal.py).
• This sandbox has under 250MB of free disk and could not install this repo's Python dependencies (pandas, pydantic, docker, pytest, etc.), so pytest test/qlib/test_workspace.py itself could not be run here. Instead, the change was validated by importing the real, unmodified workspace.py, stubbing only its heavy transitive dependencies with lightweight stdlib-only fakes, and driving QlibFBWorkspace.execute() through the same code path: it fails (AssertionError, LoadObjectError missing from the log) against the pre-fix code and passes against the fix. python3 -m py_compile was also run on both changed files with no errors.
• Note: main's own CI (the CI workflow) is currently green, so no known unrelated red checks to disclose.

Screenshots of Test Results (if appropriate):

1. Your own tests:
   Not applicable (no UI change; see "How Has This Been Tested?" above for the console-based validation performed).

Types of changes

• Fix bugs
• Add new feature
• Update documentation

Fixes #442

---

📚 Documentation preview 📚: https://RDAgent--1482.org.readthedocs.build/en/1482/

[Pujitha Paladugu (pujitha24)]

Just checking in — this has been rebased and green for a week; happy to make any changes that would help it move forward.