fix(ci): make GPU training and inference symmetric

[yuanchen8911]

Summary

Dependency: This PR depends on #577. It is opened against main like #577, so GitHub will show the #577 dedup changes in the diff until #577 merges.

Make the H100 training and inference GPU workflows symmetric, and fix the PR #579 E2E/tooling regression by pinning GoReleaser and aligning the repo toolchain to Go 1.26.2.

Motivation / Context

This is the follow-up to the GPU conformance dedup work in #577.
It removes the remaining drift between training and inference GPU CI so both workflows exercise the same core conformance checks, with only controller/gateway coverage and the inference smoke-test tail differing.

It also fixes the PR #579 E2E failure where tools/setup-tools installed GoReleaser via go install ...@latest, which drifted to v2.15.3 and required Go 1.26.2 while CI was still on Go 1.26.1.

Fixes: N/A
Related: #554, #577

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update
• Refactoring (no functional changes)
• Build/CI/tooling

Component(s) Affected

• CLI (cmd/aicr, pkg/cli)
• API server (cmd/aicrd, pkg/api, pkg/server)
• Recipe engine / data (pkg/recipe)
• Bundlers (pkg/bundler, pkg/component/*)
• Collectors / snapshotter (pkg/collector, pkg/snapshotter)
• Validator (pkg/validator)
• Core libraries (pkg/errors, pkg/k8s)
• Docs/examples (docs/, examples/)
• Other: CI workflows, chainsaw coverage, and tool/version plumbing

Implementation Notes

• Adds recipes/overlays/h100-kind-training-kubeflow.yaml and updates the training workflow to install platform: kubeflow.
• Extends training conformance coverage with robust-controller and secure-accelerator-access, and adds Kubeflow chainsaw/assert coverage for the trainer controller, webhook, and TrainJob CRD.
• Moves inference to H100 x2, adds gang-scheduling to the kind inference recipes, and removes dead deployment phase plumbing from the inference workflow.
• Aligns timeout budgets and path filters so the remaining workflow differences are the deployed platform-specific components and the inference smoke-test tail.
• Pins GoReleaser to v2.15.3 in .settings.yaml, threads that exact version through tools/setup-tools and all goreleaser-action call sites, and bumps the repo Go version references to 1.26.2.
• Follow-up cleanup is expected separately: harden goreleaser_version inputs so callers cannot drift from .settings.yaml, update stale docs that still mention older GoReleaser versions, remove the orphaned tests/chainsaw/ai-conformance/kind-training/ directory, and consider pinning the macOS brew install goreleaser path in tools/setup-tools.

Testing

unset GITLAB_TOKEN
make qualify

• make qualify passed test-coverage, lint, and e2e locally.
• make qualify failed only at scan locally because grype dir:. picked up a stale repo-root ./aicr binary built with go1.26.1; the freshly rebuilt dist/aicr_darwin_arm64_v8.0/aicr is go1.26.2.

Risk Assessment

• Medium — Touches multiple components or has broader impact
• High — Breaking change, affects critical paths, or complex rollout
• Low — Isolated change, well-tested, easy to revert

Rollout notes: This changes the inference GPU job to require H100 x2, assumes the corresponding runner class remains available, and pins release/build tooling to GoReleaser v2.15.3. Merge this after #577.

Checklist

• Tests pass locally (make test with -race)
• Linter passes (make lint)
• I did not skip/disable tests to make CI green
• I added/updated tests for new functionality
• I updated docs if user-facing behavior changed
• Changes follow existing patterns in the codebase
• Commits are cryptographically signed (git commit -S) — GPG signing info