Migrate to namespace / blacksmith

[jaimergp]

Except for GPU runners where we have no real substitute, Blacksmith and Namespace should have all the runners we need to satisfy the CPU jobs we had with Quansight- and Prefix-sponsored runners. This GH search reveals 22 feedstocks still relying on them, so I'll track the manual migration progress here.

Note that I have pre-approved access to them via:

• Remove from Cirrus, move to new providers #2008
• Grant access to Namespace and Blacksmith to repos using macOS and Windows runners on cirun #2025

PRs:

• clangdev
  • switch away from self-hosted runners on windows clangdev-feedstock#448
• flash-attn
• jaxlib
• libmagma
• llama.cpp
• llvmdev
• mongodb
• muscat-split
• nodejs
• onnxruntime
  • Bump onnxruntime to 1.24.4 and add Python 3.14 onnxruntime-feedstock#177
• openvino
• pytorch_scatter
• pytorch-cpu
  • Comparison blacksmith run pytorch-cpu-feedstock#499
  • v2.11.0 WIP pytorch-cpu-feedstock#490
• qt-webengine
• temporalio-cli
• tensorflow
• tinygrad
• torchao
• vllm
  • update to 0.19.1 vllm-feedstock#30
• vtk-m (archived)
• webkit2gtk4.1
• xformers

cc @conda-forge/clangdev, @conda-forge/flash-attn, @conda-forge/jaxlib, @conda-forge/libmagma, @conda-forge/llama-cpp, @conda-forge/llvmdev, @conda-forge/mongodb, @conda-forge/muscat-split, @conda-forge/nodejs, @conda-forge/onnxruntime, @conda-forge/openvino, @conda-forge/pytorch_scatter, @conda-forge/pytorch-cpu, @conda-forge/qt-webengine, @conda-forge/temporalio-cli, @conda-forge/tensorflow, @conda-forge/tinygrad, @conda-forge/torchao, @conda-forge/vllm, @conda-forge/vtk-m, @conda-forge/webkit2gtk4-1, @conda-forge/xformers.

[h-vetinari]

It's clear we need to replace all the linux runners; what's the status about osx/win? Presumably it'd be beneficial to migrate them, but where on the scale between "must" and "should" do we fall here?

[h-vetinari]

Ah, just saw #2025

[vicentebolea]

@jaimergp vtk-m feedstock has just been archived since Viskores is its predecessor and VTK-m development is long stopped.

[carterbox]

I planned to migrate as-needed instead of proactively. Is that ok?

[jaimergp]

Sure. The issue is mostly to notify maintainers.

[traversaro]

• llama.cpp

PR fixing this for llama.cpp is conda-forge/llama.cpp-feedstock#102 . For now we used namespace and not blacksmith as blacksmith was still affected by conda-forge/conda-smithy#2527 .