[HIP] [JIT] [gfx950] Optimize Minimax M3 scoring & top-k kernels

[ukannika]

Add fp8 MFMA sparse block scoring and top-k with fused page-table needed for pa gluon kernel.

[github-actions]

🏷️ CI Guide

Runs automatically on every PR:

• ✅ Pre-checks (submodule verification, code formatting)
• ✅ Aiter op tests (gfx942 + gfx950)
• ✅ Triton tests on MI35X (only when aiter/ops/triton/** or related paths are changed)

Extended tests (opt-in via labels):

Label	Tests
ci:gfx1250-ffm-triton	Run the five-shard gfx1250 FFM Triton test suite
ci:triton-300x	Run an additional Triton test job on MI300X in PRs; main branch always runs both MI35X and MI300X
ci:sglang	SGLang integration tests: DeepSeek-R1-MXFP4 accuracy, Qwen 3.5 accuracy
ci:atom	ATOM benchmark: DeepSeek-R1-0528, GPT-OSS-120B
ci:atom_full	ATOM accuracy suite for PR and main models from ATOM models_accuracy.json
ci:vllm	vLLM benchmark: GPT-OSS-120B, DeepSeek-R1-0528, Kimi-K2.5
ci:all	All standard extended tests (excludes ci:atom_full)

Only add ci:atom_full for FlyDSL or Triton upgrades.
Add labels via the sidebar or gh pr edit 4787 --add-label <label>

[andyluo7]

The current head cb3c7a628645 now has the full AITER test gate green, including all gfx950/MI35X shards and the final Aiter Test Gate. vLLM #52664 depends on these exact MiniMax-M3 score/top-k entry points; its focused current-head MI355X integration validation is queued now, and I will add the exact result when it completes.

Could an AITER maintainer please review #4787 for merge? This is the kernel dependency rather than a duplicate vLLM implementation.