data commands: cleaner output — drop banner, honest empty state, hide internal tables

[LukasWodka]

Two focused cleanups to the data commands' output.

1. Drop the intro banner + honest empty-state hint

The tracebloc banner block —

 tracebloc
────────────────────────────────────────
<subtitle>

— was printed at the top of every data command. It's redundant (each command's own Section/Para header already carries the context) and it hardcoded tracebloc regardless of how the CLI was launched. Removed from data list, data delete, data ingest.

For the empty data list, the hint now uses invokedName() so it echoes what the user actually typed:

 tracebloc → Datasets in <ns> (0)
──────────────────────────────────────── datasets in the cluster No datasets yet — ingest one with `tb data ingest`.
Datasets in <ns> (0)
· No datasets yet — ingest one with `tracebloc data ingest`.

(tb when launched via the tb alias, tracebloc otherwise — the same helper resources already uses.)

2. Hide the ingestor's reserved bookkeeping tables

ListDatasets ran a bare SELECT table_name … WHERE table_schema='training_test_datasets' with no filter, so the ingestor's internal bookkeeping tables leaked into data list as if they were user datasets and inflated the count. The ingestor (data-ingestors/tracebloc_ingestor/database.py) keeps two such tables and its own tests tag them "reserved":

• tracebloc_ingest_runs (RUNS_TABLE — run tracking / idempotency)
• tracebloc_ingest_meta (SALT_TABLE — pseudonymization salt)

Now filtered out at the shared ListDatasets layer, so they never surface as datasets anywhere: data list shows only real datasets, and data delete can no longer target the ingestor's bookkeeping. Reserved names live next to IngestionDatabase (kept in sync with the ingestor). The empty→nil listing contract is preserved.

Changed

• data_list.go, data_delete.go, data_ingest_local.go — drop the banner; empty data list hint via invokedName().
• push/teardown.go — reserved-table name constants next to IngestionDatabase.
• push/list.go — filterReserved applied in listDatasetsWith.

Scope

Banner removal is data commands only, by design. The shared ui.Printer.Banner method is still called by resources/resources set/cluster/doctor (being redesigned in parallel); deleting the method + remaining call sites is a small follow-up once those land.

Test plan

• go build ./..., go vet, gofmt — clean.
• go test ./internal/push/... ./internal/cli/... ./internal/ui/... — green. New: TestFilterReserved + a reserved-filtering case in TestListDatasetsWith; TestRenderDataList_Empty/_Items still pass; the ui_test.goBanner unit test is untouched (method stays).
• Verified live against a k3d env: banner gone; header reads Datasets in test0721 (N); DB had 33 tables incl. tracebloc_ingest_runs, data list showed 32 (the reserved table hidden).

🤖 Generated with Claude Code

---

Note

Low Risk
CLI output and list filtering only; behavior change is hiding internal tables and softer headers, with tests locking the filter contract.

Overview
Data command UX drops the redundant tracebloc intro banner on data list, data delete, and data ingest (replaced with a leading newline where needed); ingest/delete keep their explanatory paragraphs. Empty data list now suggests ingest using invokedName() so the hint matches how the binary was launched (tb vs tracebloc).

Dataset listing filters ingestor-internal MySQL tables (tracebloc_ingest_runs, tracebloc_ingest_meta) in ListDatasets via filterReserved, with constants documented next to IngestionDatabase. That fixes inflated counts and prevents those names from appearing as deletable datasets (delete already resolves targets through the same list). Tests cover filtering and the empty→nil slice contract.

^{Reviewed by Cursor Bugbot for commit 0ed3404. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

/fr-pass — verified live on dev: data list shows no banner, header "Datasets in test0721 (32)", and the reserved tracebloc_ingest_* bookkeeping tables are filtered out of the count.