GDN packed sequence support landing in Nemo containers

Currently tracking packed sequence support for GDN (Gated Delta Net), specifically PRs #2644 and #2645 (e.g., https://github.com/NVIDIA/Megatron-LM/pull/2644) in megatron-lm being merged to mainline and the Q2 roadmap lists it as targeting the 26.06 release. Some follow up questions:

1. The fix landed on main on May 14, after the core_v0.17.0 tag (April 14). Will it be cherry-picked into a 0.17.x patch release, or will it ship as part of 0.18?
2. Is 26.06 / the June NeMo container expected to include 0.17 or 0.18?
3.  from the PR: it requires cudnn>=9.19.0 for Qwen3.5 THD training on Hopper to avoid grad NaN. Can we confirm which NeMo container will have both cudann >= 9.19.0 and the fix?