docs(concepts): update per-backend terminology map

[Copilot]

The backend hierarchy table was stale: it listed the removed pandas backend, omitted supported backends, and used inaccurate terminology for several systems.

• Coverage

  • Added Athena, Databricks, Exasol, Materialize, and SingleStoreDB.
  • Removed pandas.

• Terminology

  • Corrected verified native hierarchy names: BigQuery datasets, PySpark catalogs/databases, and Oracle's database/schema levels.
  • Replaced hierarchy levels a backend does not have with N/A.
  • Removed the Druid datasource terms, which do not represent catalog/database hierarchy; Druid's database level is its SQL schema.

[deepyaman]

I verified every row in this table against the backend implementations on main rather than taking the description's "verified" claim at face value. Summary: the diff looks alarming (−20/+24) but that's almost entirely because the column widths changed, forcing a rewrite of all 18 original rows. Substantively it's two real corrections, one wrong row replaced, one removal, and five additions.

Completeness ✅

pyproject.toml registers exactly 22 backend entry points, and the new table has exactly those 22 rows — nothing missing, nothing invented. pandas is genuinely gone (no ibis/backends/pandas, no entry point). The pre-existing sqlite/snowflake sort-order bug is also fixed.

Substantive changes, with evidence

Change	Verdict	Evidence
bigquery database → dataset	correct	bigquery/__init__.py:307-308 — database = self._session_dataset.dataset_id, catalog = ....project; do_connect takes dataset_id
pyspark database|schema → catalog|database	correct	pyspark/__init__.py:256-263 — SELECT CURRENT_DATABASE() / SELECT CURRENT_CATALOG()
druid dataSourceType|dataSource removed	correct to remove — a dataSource is a table, and dataSourceType isn't a hierarchy level at all	druid/__init__.py:69
athena data catalog | database	correct	athena/__init__.py:308 list_data_catalogs(), then list_databases(catalog_name=...)
databricks catalog | schema	correct	databricks/__init__.py:359-366 — cur.schemas(catalog_name=...) → TABLE_SCHEM
exasol N/A | schema	correct	exasol/__init__.py:443 SELECT CURRENT_SCHEMA; :488 rejects catalog
materialize database | schema	correct	materialize/__init__.py:20 — class Backend(PostgresBackend)
singlestoredb N/A | database	correct	singlestoredb/__init__.py:190 SELECT DATABASE(); matches the "SingleStoreDB doesn't support catalogs" comment in test_client.py
blank → N/A for clickhouse/impala/mysql/polars/sqlite	correct, cosmetic	none of these define current_catalog

Two cells I've corrected in b05e574

oracle: N/A → database (Catalog). Oracle was the only backend this PR marked N/A that actually implements current_catalog — oracle/__init__.py:239-242 returns global_name. Natively Oracle does have a database → schema hierarchy, and this table documents native terminology, so database is the accurate value. (The database → schema fix in the Database column was right and I've kept it: oracle/__init__.py:245-250 returns the user, with the comment "databases correspond to users".)

The counterargument for N/A is that Ibis can't traverse it — list_databases rejects a catalog argument (oracle/__init__.py:328), and Oracle has no db.schema.table qualification without DB links. Happy to be overruled here, but I think the table's stated purpose ("the terminology used by each backend for the two levels of hierarchy") favors describing the backend rather than Ibis's support level.

druid: N/A → schema (Database). druid/__init__.py:64-66 returns the hardcoded "druid", citing the INFORMATION_SCHEMA.SCHEMATA docs — Druid SQL calls that level a schema. N/A isn't wrong in spirit (Druid has no list_databases, and test_client.py marks it notyet for "the common notion of a database"), but schema is more informative and consistent with how the rest of the column is filled in. This is the softer of my two changes — revert it if you'd rather the column reflect Ibis support.

I also padded the singlestoredb row, which was one character wider than the separator. For the record this would never have failed CI: identify tags .qmd as {'text', 'file', 'non-executable'}, not markdown, so the prettier pre-commit hook skips this file entirely.

🤖 Generated with Claude Code

---

_{Follow-up, not blocking this PR: once the chdb backend lands (#12059), this table will need a chdb | N/A | database row — chdb defines no current_catalog and mirrors ClickHouse's single-level hierarchy. Flagging it so the table doesn't immediately go stale again. Whichever of the two merges second should pick it up; I'm happy to add the row on that side.}