[feat][DCP] Enable DCP with fp8 MTP

[yitingw1]

Motivation

DCP + MTP (speculative decode) previously required bf16 KV cache — the round-robin CP (cprr) MLA kernel had no fp8 variant, so ATOM raised NotImplementedError for fp8. The ROCm/aiter#4521 now ships fp8 cp-round-robin kernels (gqa16/gqa32, qseqlen4, plain + lse), so fp8 KV cache can be combined with MTP under DCP — cutting per-GPU KV memory further while keeping speculative decode.

Need ROCm/aiter#4521 merged first.

Test

8*MI355 DeepSeek-R1, -tp8, --method mtp, gsm8k 5-shot (flexible / strict), 0 crash:

KV dtype	num_spec	dcp	gsm8k
fp8	1	8	0.9560 / 0.9545 (full 1319)
fp8	2	8	0.9553 / 0.9560 (full 1319)
fp8	3	8	0.9538 / 0.9507 (full 1319)
fp8	3	4	0.95 / 0.95 (limit 100)
fp8	3	2	0.94 / 0.94 (limit 100)

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[github-actions]

🏷️ CI Guide

Runs automatically on every eligible PR before approval:

• ✅ Pre Checkin: Black, Ruff, catalog schema validation, non-GPU unit tests

Heavy model tests:

• ✅ Run after the PR is approved and Pre Checkin passes
• ✅ Run immediately when an approval review is submitted
• ✅ Can be requested before approval with labels

Label	Tests
ci:full	Run all heavy PR model tests: native ATOM, vLLM, and SGLang
ci:atom	Run native ATOM model accuracy tests
ci:vllm	Run ATOM vLLM OOT model accuracy tests
ci:sglang	Run ATOM SGLang model accuracy tests

Heavy jobs are skipped when the PR is not approved and no matching ci:* label is present.
Add labels via the sidebar or gh pr edit 1781 --add-label <label>