build: AUT-1613 bump transformer-engine to release_v2.18 tip

[svcnemo-autobot]

Background

The r0.6.0 release branch pinned Transformer Engine 2.17.1. Update it to the tip of TE's release_v2.18 branch and retain its build requirements.

What changed

• Pin Transformer Engine to 27486e03cfc1fa41f6932dcecdc47c71c47eac3e.
• Restore TE 2.18's nvidia-cudnn-frontend>=1.25.0 dependency metadata and build group.
• Update matching lockfile source, metadata, and dependency entries.

Details

flowchart LR
  A[r0.6.0] --> B[TE release_v2.18 tip]
  B --> C[cuDNN frontend build dependency]
  C --> D[uv.lock]

Loading

The Megatron-LM gitlink remains unchanged.

Tested

• git diff --check
• Verified the TE SHA/version and absence of stale 2.17.1 entries
• Verified every restored cuDNN frontend entry against the prior known-good TE 2.18 lock
• Dependency-complete image build and required CI in progress

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[svcnvidia-nemo-ci]

/ok to test 58984e8

[svcnvidia-nemo-ci]

/ok to test adbfecc

[svcnemo-autobot]

CI infrastructure note: all product and GPU jobs passed on attempts 2 and 3. The only failing job is Nemo_CICD_Test; both attempts failed while gh run view queried the jobs API and received HTTP 502: Server Error. The bounded automated retry is exhausted. Could a maintainer rerun the failed job/workflow after the GitHub API recovers?