Add offline WGMM variant evaluator and burst-aware scheduling adjustment

[334456777]

Motivation

• Add an offline evaluation harness to compare WGMM prediction variants using release data.
• Improve scheduler robustness by making frequency decisions aware of recent publication bursts.

Description

• Add tools/eval_wgmm_variants.py which implements predict_baseline, predict_hazard, and predict_cooldown predictors plus an evaluate harness to run randomized replay experiments against data/mtime.txt and data/miss_history.txt.
• Introduce MIN_BURST_EVENTS and compute_burst_adjustment in wgmm_monitor/wgmm/scheduler.py to compute a burst-intensity ratio from recent publishes.
• Apply the burst adjustment to the scheduler's relative_score inside decide_next_frequency so the mapping to check intervals is dampened or amplified based on recent publishing intensity.

Testing

• Ran the repository test suite with pytest -q and the tests completed successfully.

---

Codex Task