[chuiwenwei]

Summary

LanceDB has no unique constraint, so a write path that bypasses the merge
helper can leave two rows for one (app_id, project_id, id). The next
cascade write may pick one arbitrarily and search silently loses the
other. Add an opt-in everos cascade status --duplicates scan that counts
extra rows per logical key in every business table.

The scan is O(N) per table, so it stays off the default status path.

Closes #479.

Area

• Architecture method
• Benchmark
• Use case
• Documentation
• Developer experience
• CI, build, or release

Verification

make lint        passed
pytest tests/unit/test_infra/test_index_diagnostics.py \
       tests/unit/test_entrypoints/test_cli/test_cascade_command.py   9 passed

The full unit suite will run in CI; the local run was stopped under host
load after the targeted tests passed.

Checklist

• I kept the change scoped to the relevant area.
• I am opening this from a separate branch, not pushing directly to main.
• I updated docs, examples, or setup notes when behavior changed.
• I added or updated tests when the change affects behavior.
• I did not commit secrets, .env files, dependency folders, or generated output.
• Active relative links in Markdown files resolve.

Notes for Reviewers

• The flag is opt-in so cascade status stays cheap for normal use.
• The scan is backend-neutral; Milvus enforces its own primary key and
  should report zero.
• docs/cli.md and docs/cascade_runbook.md describe the new flag.

By submitting this pull request, I agree that my contribution is licensed under
the Apache License 2.0.