docs: replace single-bug Known Issue with full Serving status (3 upstream bugs filed: #42741, #42769, #42777)

docs: add Known Issue section for vLLM x transformers compress_ratios regression (#42741 upstream)

fix: restore compress_ratios + num_hash_layers + rope_scaling (vLLM 0.21.0 deepseek_v4 model expects pre-normalization fields)

DeepSeek-V4-Flash-TQ3-native Gate C publish

initial commit