ci(release): drop redundant E2E from the orchestrate build path

[joshua-temple]

Problem

The orchestrate workflow calls the hand-written build-cli.yaml reusable workflow, which had an e2e-tests job running the full act+gitea e2e suite a second time on every release. That coverage is already enforced as the required PR gate (e2e.yaml, sharded, on pull_request/merge_group), so re-running it during the release path was redundant compute with no additional confidence.

Fix

• Removed the e2e-tests job from .github/workflows/build-cli.yaml.
• Repointed the workflow_call output from jobs.e2e-tests.result to jobs.build.result, since build is now the only job.
• Added a one-line note that end-to-end coverage lives in the PR gate (e2e.yaml), not here.

build-cli.yaml is hand-written (no AUTO-GENERATED header) and is not touched by the generator. orchestrate.yaml (generated) still calls ./.github/workflows/build-cli.yaml unchanged and reads needs.build-cli.outputs.result, which now reflects the build job instead of the old e2e job. e2e.yaml (the PR gate) is untouched. The fleet still validates end-to-end behavior on live GitHub at the rc gate, independent of this path.

Verification

• go build ./... clean.
• cascade verify --config .github/manifest.yaml reports no drift (3 files, no drift) - confirms this hand-written change doesn't desync the generated workflows.
• grep -n e2e-tests .github/workflows/build-cli.yaml returns nothing; grep -n jobs.build.result .github/workflows/build-cli.yaml matches the new output line.
• Docs checked (docs/src/content/docs/workflows.md, coverage-matrix.md, architecture.md, getting-started.md, README.md): none claim the release/orchestrate path runs e2e, so no doc updates were needed.
• actionlint on the file reports a pre-existing false positive on the workflow_call output expression (jobs.<id>.result isn't recognized by its type checker in this shape); this was already present against jobs.e2e-tests.result before this change and is unrelated to it.