sync local: enforce_mix sampler, tokenizer v2 (mask fix + 35/30/35), mix optional docs, repair scripts

docs: add official comprehensive data audit report (145.93GB/105.1B tokens - 100% PASS)

Add Step 4 30B-50B+ Kaggle data collection suite & 20-pillar knowledge matrix

feat: SDPA FlashAttention, language tags conditioning, probe safety, slim checkpoints, and streaming clean

init: 28L MoE 241M audit-fixed, smoke pass