feat: allow online FP8 quantization with W4A16 dense FFN

[HongminTan]

No description provided.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[rtp-llm-review-bot]

PR #1454 评审：未发现阻塞项，等待人工审核

• 标题：feat: allow online FP8 quantization with W4A16 dense FFN（@HongminTan，open）
• 评审版本：ade358e46ca1

无阻塞项

非阻塞发现

• [P2] support 的在线 FP8 方法白名单遗漏 "FP8"，会错误拒绝合法的在线 per-tensor FP8 配置 — rtp_llm/models_py/kernels/cuda/w4a16_sm120.py:39
• [P2] quantize_weight 以 float32 权重块调用 rotate，若 hadamard op 仅支持 bf16 会失败 — rtp_llm/models_py/kernels/cuda/w4a16_sm120.py:119

已确认修复

• [P2] support 白名单包含无量化权重类的 "FP8" 方法，启用时会静默退化为 BF16 — 当前 head 第 39–43 行白名单已移除 "FP8" 和 "FP8_PER_CHANNEL_QUARK"，仅保留 FP8_DYNAMIC_PER_TENSOR/FP8_PER_BLOCK/FP8_PER_CHANNEL_COMPRESSED，发现所指的 "FP8" 方法放行路径已被删除。

---

_{自动代码评审 · head ade358e46ca1 · 未发现阻塞项，等待人工审核}