Postgres: `postgres.indexes` entry without a `name` fails with "zero-length delimited identifier"

[ihistand]

Found during the 1.4.0 acceptance pass (#17), real-user flow: published @sqlanvil/cli@1.4.0 against a local Postgres, no repo checkout.

Repro

config {
  type: "table",
  postgres: { indexes: [ { columns: ["id"], unique: true } ] }
}
SELECT 1 AS id

sqlanvil run →

Dataset creation failed: sqlanvil.iso_idx [table]
postgres error: Error executing postgres query: zero-length delimited identifier at or near """"

Cause

An index config entry that omits name generates CREATE [UNIQUE] INDEX "" ON ... — an empty ("") index identifier — which Postgres rejects. No index name is auto-generated.

Confirmation

Adding an explicit name works and the index is created correctly:

postgres: { indexes: [ { name: "iso_idx_named_id_uq", columns: ["id"], unique: true } ] }

→ Table created, \di shows iso_idx_named_id_uq.

fillfactor alone and description/columns (COMMENTs) alone are unaffected — the bug is specific to index entries without a name.

Impact

Blocker-level for a common case: a table with any unnamed postgres.indexes entry fails to create. The MVP checklist's own example (mvp_user_test_checklist.md §4) omits the name, so a doc-following user hits a hard failure.

Fix direction

Auto-generate an index name when name is omitted (e.g. <table>_<col1>_<col2>_idx, matching Postgres's own default naming, deduped/truncated to 63 chars), in the Postgres execution-SQL index generator. Add a unit test for the no-name case and an integration test in postgres.spec.