glm5next: avoid soft_max gridDim.y overflow in the indexer

[AIalliAI]

Overview

Reshape k-pool indexer gate logits before ggml_soft_max so n_new_max does not map to gridDim.y, which is capped at 65535 on CUDA. During k-pool rebuild at long context (n_kv >= 262144, kpool = 4), n_new_max reaches 65538 and the launch aborts with SOFT_MAX failed.

Fold n_new_max and n_stream into ne1 via ggml_reshape_2d, then reshape back to 4D for downstream mul / sum_rows. Same pattern as qwen4exp.cpp.

Stacked on ggml-org#27754.

Additional information

Related: ggml-org#27773, timkhronos#11, ggml-org#27901

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES - assisted implementation and review; I understand the change and can maintain it.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.