ci(e2e): reuse prebuilt CLI and gateway artifacts

[elezar]

Summary

Reuse the architecture-matched CLI and gateway binaries already produced by the branch build jobs instead of rebuilding them inside each E2E job. This PR is intended to verify whether removing those duplicate builds reduces E2E execution time.

Related Issue

None; follow-up to the merge queue CI work.

Changes

• Download and configure the prebuilt CLI for core, GPU, and Kubernetes E2E jobs
• Download and configure the prebuilt gateway for Docker, Podman, and GPU E2E jobs
• Preserve local fallback builds when no prebuilt binary is supplied
• Use deterministic artifact paths and concise architecture-specific artifact names

Testing

• mise run pre-commit passes in ~/.dotfiles/devenvs/openshell
• Unit tests included in the pre-commit task pass
• E2E tests pass in CI via the test:e2e label

E2E Timing Results

The artifact-reuse run completed in 24m32s, compared with 40m20s for the recent successful same-target baseline. Both PRs targeted main@3dee5570. This is a 15m48s (39%) reduction in workflow wall time.

Common lane	Baseline	Artifact reuse	Improvement
Rust Docker	8m53s	4m24s	4m29s
Python	7m14s	3m09s	4m05s
MCP	7m12s	3m12s	4m00s
Podman Ubuntu 24.04	18m00s	10m40s	7m20s
Podman Ubuntu 26.04	17m40s	10m51s	6m49s
Kubernetes sidecar	9m34s	8m04s	1m30s
Kubernetes v1alpha1	10m23s	8m28s	1m55s
Kubernetes v1beta1	9m56s	8m27s	1m29s

Across the eight comparable lanes, aggregate job runtime fell from 88m52s to 57m15s, a 35.6% reduction. Artifact preparation did not delay fan-out: E2E started after 13m36s, versus 13m43s in the baseline. The CLI build replaced the gateway image as the build-phase critical path, but remained within the existing preparation window.

The baseline PR added an extra rust-vm lane, so whole-workflow topology is not identical. That lane was not the baseline critical path; every common lane improved, making the per-lane comparison the stronger signal.

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)

[github-actions]

Label test:e2e applied for 8ee3bbb. Open the existing run and click Re-run all jobs to execute with the label set. The run will execute the standard E2E suite after building the required gateway and supervisor images once. The matching required CI gate status on this PR will flip green automatically once the run finishes.

[krishicks]

Similar to #1910. I found value in making the test output understandable because the log doesn't also include the cargo compile output. I closed that one because I thought a much bigger change would be better, but given that will take a while as we're still at the deciding phase, I'm happy to see this.