fix(tests): e2e-journey Step 4 uses offline data validate instead of push --dry-run

[saadqbal]

What

Closes #317. Follow-up to #316 (which fixed the create_cluster exit-127 regression that was stopping the E2E last-mile journey before Step 4 ever ran).

With #316 merged, the E2E job reaches Step 4 for the first time with a working guard() and fails:

── Step 4: dataset push --dry-run smoke ──
Error: table name is required (set --table)
exit code 2

Why the old Step 4 can't work here

dataset push --dry-run is no longer an offline check. The current CLI (internal/cli/data.go) runs cluster discovery and a Bound-PVC check (resolveClusterTarget(..., needPVC=true)) and a destination-table check before the dry-run stop. This harness's credential-free stub creates no PVC, so dataset push --dry-run can't pass against it regardless of which flags we add.

Fix

Switch Step 4 to tracebloc data validate <spec> — the CLI's purpose-built offline schema check (validates a spec against the embedded ingest.v1.json, never touches the cluster). That's exactly the "offline-validatable, no creds, no reachable platform" smoke the step was always documented to be.

• Committed spec mirrors the CLI's own known-good fixture (testdata/smoke/valid-image-classification.yaml); the /data/shared paths are schema-only (validate never reads them).
• Asserts both directions — valid spec passes (exit 0), invalid spec (missing required intent) is rejected (exit 2) — so a green run isn't vacuous, same discipline as the guard() self-test.

Verification

Built the actual released binary tracebloc/cli v0.6.0 (what releases/latest resolves to; the version the E2E installs) and ran both files:

ingest.yaml: ok                                    exit=0
ingest-invalid.yaml: schema validation failed …    exit=2   (missing property 'intent')

bash -n clean.

🤖 Generated with Claude Code

---

Note

Low Risk
Test-only change to the E2E journey script; no production code or runtime behavior is modified.

Overview
Step 4 of e2e-journey.sh no longer runs tracebloc dataset push … --dry-run on a sample CSV. It now exercises tracebloc data validate against a generated IngestConfig YAML (aligned with the CLI smoke fixture), which stays credential-free and does not need cluster discovery or a Bound PVC.

The step asserts both outcomes: a complete spec must exit 0, and a spec missing required intent must fail non-zero so a broken validator cannot pass vacuously. Header comments for the journey are updated to describe this offline schema smoke instead of dataset push dry-run.

^{Reviewed by Cursor Bugbot for commit 6db20af. Bugbot is set up for automated code reviews on this repo. Configure here.}