fix: qlib backtests fail instantly under recent MLflow (filestore gated)

[chrisbatso]

Problem

Recent MLflow releases put the filesystem tracking backend (./mlruns) into maintenance mode and refuse to use it unless MLFLOW_ALLOW_FILE_STORE=true is set:

MlflowException: The filesystem tracking backend (e.g., './mlruns') is in maintenance mode
and will not receive further updates. ... If the filesystem backend is required for your
workflow, set MLFLOW_ALLOW_FILE_STORE=true to opt out of this exception.

The qlib Docker image (rdagent/scenarios/qlib/docker/Dockerfile) installs MLflow unpinned as a qlib dependency, and qlib's recorder is built on the ./mlruns filestore — so every qrun inside the container now dies within seconds. From the R&D loop's perspective the failure is quiet: the running step returns almost instantly with no result file, and every experiment is judged on a missing backtest.

Fix

Set ENV MLFLOW_ALLOW_FILE_STORE=true in the image. One line, no behavior change for older MLflow versions (the variable is simply ignored), restores qrun under new ones.

Testing

Reproduced with a freshly built image (Docker 29, image built 2026-08-29): qrun conf_baseline.yaml raised the MlflowException immediately; after rebuilding with this ENV, baseline and combined-factor backtests complete and read_exp_res.py extracts metrics and portfolio artifacts normally. Verified across multiple full 15-loop fin_factor runs.

🤖 Generated with Claude Code

---

📚 Documentation preview 📚: https://RDAgent--1464.org.readthedocs.build/en/1464/