Task: Fix the artefact index emission and the tags ON CONFLICT

(Describe what user-visible-or-internal change this task produces.)

• A staging table emits its key index once, however many artefact columns it declares.
• ores_dq_tags_artefact_tbl carries a unique index on (tenant_id, dataset_id, name), so the seed's on conflict clause has an arbiter.
• A declared artefact index is unique only when it says :unique: true.
• A check fails on an unterminated block comment and on a repeated index in one file, and it runs on the pull request as well as in Continuous Linux.

1. Confirm the two defects. The five population-function comments were already fixed by PR #2227; a PostgreSQL-aware scan proves the current tree is balanced. The tags on conflict and the repeated indexes are live.
2. Fix the index emission at source. Move the artefact key index out of the artefact-columns loop in ores.sql.schema.domain_entity_artefact_create, and teach the * Artefact indexes section a :unique: flag.
3. Declare the arbiter. Add the unique tag_identity artefact index to ores.assets.tag, which owns the staging table the seed writes.
4. Regenerate and narrow the diff. Regenerate the SQL facet tree-wide, then keep only the ten dq_*_artefact_create.sql files this template can write and revert the drift the run materialized elsewhere.
5. Pin both behaviours in tests. The key index count is asserted exactly, which the old presence assertion did not do, and the unique flag is asserted both ways.
6. Encode the classes as a check. check_sql_hygiene.py reads the tree for unterminated block comments and repeated indexes; validate_schemas.sh and the codegen-drift job both call it.

Manual QA scenarios (scaffolded via compass add test_scenario) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

Both defects are fixed at source and both classes are gated.

The index emission. The key index was inside the artefact_columns loop, so a table rendered one copy per artefact column: 245 lines of duplicate index across ten staging tables, worst dq_currencies_artefact_iso_code_idx at seventeen copies. It is now emitted once, under artefact_key_column.

The missing arbiter. ores_dq_tags_artefact_tbl had no unique index at all, so the seed's on conflict (tenant_id, dataset_id, name) had nothing to infer and would have failed at run time. The * Artefact indexes section gained a :unique: flag and the tag model declares the index.

The five comments were already fixed. PR #2227 closed them. The scan written here confirms the tree is balanced and would have caught the original five, which is the evidence that the fix holds.

Two findings recorded, not fixed here. Regenerating the SQL facet tree-wide materialized generated SQL that main has never committed – marketdata, refdata, synthetic and a whole workspace table with its trigger and drops. They are reverted here and belong to their own tasks. And regenerate_physical_space_inventories.py --check reports three stale facet inventories (ores.org, ores.cpp.org, ores.cpp.http-route.org); it reproduces on a stashed clean tree, the staleness is a row reordering, and the gate is in no workflow, so it is recorded rather than swept into this change.