Keep CSR column ids int32 when the edge index is int32

[dsaini2-sc]

Summary

Since #786 the partitioners hand DistDataset an int32 edge index whenever node ids fit, but the compact CSR build (#768) still widened the column ids to int64. This PR keeps them int32, which GiGL's patched GLT samplers (#761) read directly. Peak memory for the CSR conversion drops from ~2x one int64 edge array to ~1.5x, and the stored column ids are half the size. int64 inputs are unchanged.

Approach

• build_csr_from_coo allocates indices with the width of col; indptr stays int64.

User-facing changes

• Topology.indices on the compact path is int32 for int32 edge indices. Graphs with edge ids, weights or features still go through GLT's init_graph and stay int64.
• This needs the GLT patches from Apply and verify local patches on the pinned GraphLearn-Torch build #761 in the image. The base images are rebuilt and the pins bumped in this PR by the build-base-docker-images workflow.

Verification

• csr_test, compact_topology_test (adds int32 CSR/CSC sampling parity against GLT's own Topology), glt_int32_indices_test: 64 tests pass.
• distributed_neighborloader_test (36, one pre-existing skip), distributed_dataset_test (15), distributed_partitioner_test (34): pass, with real spawned loaders.
• Full unit suite (1274 tests, runtime shape checks on): passes except 3 config_populator tests that need CI bucket access the dev machine lacks.
• ruff, ty, dprint: clean.

[github-actions]

GiGL Automation

@ 18:48:26UTC : Starting to build base images for CUDA and CPU.
This may take a while, please be patient.
The images will be pushed to the GCP Artifact Registry.
Once done, the workflow will update the gigl/dep_vars.env file with the new image names.

[github-actions]

GiGL Automation

@ 18:58:18UTC : Built and pushed new images:

• CUDA base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-cuda-base:6caecb4ed5a703d7a30b27a37f65ae10747e23af.114.1
• CPU base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-cpu-base:6caecb4ed5a703d7a30b27a37f65ae10747e23af.114.1
• Dataflow base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-dataflow-base:6caecb4ed5a703d7a30b27a37f65ae10747e23af.114.1
• Builder image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/gigl-base-images/gigl-builder:6caecb4ed5a703d7a30b27a37f65ae10747e23af.114.1

Updated gigl/dep_vars.env with new image names.
Updated .github/cloud_builder/run_command_on_active_checkout.yaml with new builder image.

[zfan3-sc]

approve to unblock. lgtm once Kyle's comment is addressed

[github-actions]

GiGL Automation

@ 05:04:05UTC : Starting to build base images for CUDA and CPU.
This may take a while, please be patient.
The images will be pushed to the GCP Artifact Registry.
Once done, the workflow will update the gigl/dep_vars.env file with the new image names.

[github-actions]

GiGL Automation

@ 05:12:21UTC : Built and pushed new images:

• CUDA base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-cuda-base:86b4bc2e01ded1e5181c56fb7f85015833676308.115.1
• CPU base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-cpu-base:86b4bc2e01ded1e5181c56fb7f85015833676308.115.1
• Dataflow base image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/public-gigl/gigl-dataflow-base:86b4bc2e01ded1e5181c56fb7f85015833676308.115.1
• Builder image: us-central1-docker.pkg.dev/external-snap-ci-github-gigl/gigl-base-images/gigl-builder:86b4bc2e01ded1e5181c56fb7f85015833676308.115.1

Updated gigl/dep_vars.env with new image names.
Updated .github/cloud_builder/run_command_on_active_checkout.yaml with new builder image.