ZhiyuanChen/DanLing: Development Build

Commits

• improve runner config defaults (Zhiyuan Chen)
• honor runtime config knobs (Zhiyuan Chen)
• fix step-mode progress logging (Zhiyuan Chen)
• add rank context for warnings (Zhiyuan Chen)
• classify storage failures (Zhiyuan Chen)
• add lifecycle summaries (Zhiyuan Chen)
• report profiler artifacts (Zhiyuan Chen)
• shorten memory metric names (Zhiyuan Chen)
• harden training safety and telemetry (Zhiyuan Chen)
• add checkpoint and FSDP module policies (Zhiyuan Chen)
• standardize seeds and runner coverage (Zhiyuan Chen)
• add tensor loss parallel context (Zhiyuan Chen)
• reduce dtensor losses locally (Zhiyuan Chen)
• keep checkpoint failures recoverable (Zhiyuan Chen)
• cover distributed precision edges (Zhiyuan Chen)