[Feature] JIT activation and update skills (by codex)

[DarkSharpness]

Motivation

Modifications

Accuracy Tests

Speed Tests and Profiling

(TL;DR: Performance gain is mostly from PDL and vectorization)

H200:

B200:

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[DarkSharpness]

/tag-and-rerun-ci

[gemini-code-assist]

Code Review

This pull request introduces JIT-compiled activation kernels (SiLU, GELU, and GELU-Tanh with multiplication) to replace AOT implementations on CUDA platforms. The changes include the core CUDA kernel implementation, Python wrappers, unit tests, and benchmarks. Documentation is also updated to include guidance on JIT kernel development and the new PDL (Primary-to-Secondary Dependency Link) feature. Review feedback identifies a missing import for the HIP platform in the multimodal runtime and points out an incorrect output shape registration in the custom operator metadata for activations.

[Fridge003]

Hi @DarkSharpness, this PR was reverted due to this failure https://github.com/sgl-project/sglang/actions/runs/23958698449/job/69895069178?pr=21913
Please help relanding later