fix(tests): e2e-journey Step 4 is stale — dataset push --dry-run needs a Bound PVC + new flags

[saadqbal]

Problem

Once #316 fixed the create_cluster exit-127 regression, the E2E last-mile journey (scripts/tests/e2e-journey.sh) reaches Step 4 for the first time with a working guard(). Step 4 fails:

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

Two independent reasons the step is stale vs. the current tracebloc CLI:

1. dataset push --dry-run now requires --table + --category (+ label/schema resolution).
2. More fundamentally, the CLI's --dry-run is no longer offline — it runs cluster discovery and a Bound-PVC check (resolveClusterTarget(..., needPVC=true)) before the dry-run stop. The harness stub creates no PVC, so dataset push --dry-run cannot pass against it regardless of flags.

The step's own comment still claims it's "offline-validatable; no creds, no reachable platform" — which is no longer true of dataset push --dry-run, but is exactly what the offline tracebloc data validate <spec.yaml> does (schema-validates a spec against the embedded ingest.v1.json, no cluster).

Fix

Rewrite Step 4 to validate a committed minimal ingest.yaml with tracebloc data validate (mirroring the CLI's own known-good smoke fixture testdata/smoke/valid-image-classification.yaml). Restores the intended offline/credential-free smoke; no PVC needed.

Follow-up to #316 / #310.