fix(validate): skip gcp-gce revalidation (composite vs component SKU mismatch)

[sofq]

Summary

• Adds a per-shard `SKIP_REVALIDATION` dict in `pipeline/validate/driver.py`. Listed shards short-circuit with `exit="skipped"` and return 0.
• Initial entry: gcp-gce. Ingest synthesizes machine totals from per-vCPU and per-GiB component prices (`pipeline/ingest/gcp_gce.py:140`); the validator reads the raw component `unitPrice` for the matching SKU id, so the two are not comparable. Result: 20/20 false-positive drift records on every run (Catalog drift in gcp-gce #62).
• The other open drift issues (Catalog drift in azure-postgres #56-61) need live-API verification before action — not addressed here.

Test plan

• `uv run pytest pipeline/tests/test_validate_driver.py` — new `test_driver_skips_listed_shard` passes; existing tests unchanged.
• `uv run pytest pipeline/tests/test_validate_gcp.py pipeline/tests/test_validate_sampler.py` — green.
• Next data-validate run for `gcp-gce` returns success → workflow auto-closes Catalog drift in gcp-gce #62.

Follow-up

A proper fix is a validation hint sidecar (ingest writes `expected_upstream_amount` per SKU; validator just compares those). Tracking issue to file once this lands.