fix(cli): record a submission when no autograder is configured

[rongxin-liu]

Summary

PR #458 made the autograde workflow's grade and set-latest jobs skip entirely when an assignment has no autograder configured (no declarative tests, no per-assignment bundle, no classroom-default autograder.py). That correctly stopped spending Actions minutes, but it also stopped the side effect that records a submission: the submit/* GitHub Release. The submit/* tag was still minted in the setup job, but the teacher submissions dashboard reads Releases (via latestSubmitReleaseAndCount), not tags — so a student who pushed to a no-autograder assignment silently vanished from the teacher's view and it read like a bug.

This restores the "records" half while keeping #458's cost win. The grade job now runs for the no-autograder case so runner.py's existing vacuous-pass path (Finalizer.no_autograder()) synthesizes a result.json scoring 0/0 with status=success, which the existing "Publish submission release" step publishes as the submit/* release. Language-toolchain setup (Python/Node/Java/Go/Rust + apt) is gated off no-autograder, so the recording path provisions no toolchain it won't use — it needs only system python3.

This is forward-compatible with the "milestone tags trigger; submit/* records" invariant reaffirmed in the draft PR #531: submit/* stays the canonical record channel.

Changes

• grade.if now gates only on the acceptance commit; the no-autograder clause is dropped so the job runs and records the submission.
• The five toolchain setup steps and the apt step now also carry && needs.setup.outputs.no-autograder != 'true', preserving the feat(cli): skip autograde grade job when no autograder is configured #458 minute savings.
• Removed the setup-job "Autograde-skipped commit status" step — the grade job's always() "Post commit status" step already posts classroom50/autograde=success with the runner's "no autograder configured" summary, so the status-polling channel is at parity with no double-post.
• set-latest drops its no-autograder guard so the recorded submission advances the latest pointer.
• Updated the init_skeleton_test.go parity assertions (the pinned grade.if string and the six toolchain if: strings) and a stale comment in test_inline_validator.py. No runner.py logic change — its no-autograder synthesis was already correct and covered.

Verification

• go test ./... in cli/gh-teacher passes, including TestSkeletonFiles_AutogradeRunner; go vet and gofmt clean.
• python3 -m pytest cli/gh-teacher/autograders_tests — 358 pass, including test_no_bundle_and_no_default_synthesizes_vacuous_pass, which pins the end-to-end 0/0-success recording (result.json, release body, status=success).
• Workflow YAML parses; jobs remain setup/grade/set-latest (no job added).
• No web changes needed: the dashboard surfaces the release automatically once it exists again.