GeezerVison

Fixes #4657

Problem

targetName returned "?column?" for a result column with no alias that is not a column reference or a function call. The Go generator uses a non-empty name as written, so it emitted var ?column? ... and sqlc generate failed with illegal character U+003F '?'.

The PostgreSQL and SQLite engines name such a column by position (column_1).

Change

targetName (internal/core/analyzer/projection.go) returns an empty name for such a target. The Go generator already falls back to column_N for an empty name (columnName in internal/codegen/golang/result.go). ?column? was the only place in non-test Go code that produced the string.

Tests

New end-to-end case: duckdb_unnamed_columns/duckdb (sqlc generate, output committed). It covers an unaliased CAST, a || expression, two unnamed columns, and a mix of named and unnamed columns. It fails without the change (illegal character U+003F '?') and passes with it.

go test ./internal/endtoend -run TestReplay/base: 971 pass, 58 skip, 0 fail before; 972 pass, 58 skip, 0 fail after. The only difference is the new case. No existing golden contained ?column?.
go test ./internal/engine/... ./internal/core/... ./internal/compiler/... ./internal/codegen/... ./internal/cmd/... ./internal/metadata/... ./internal/sql/...: all ok.

Not run: the managed-db context, --tags=examples, and the internal/goldeneye checks.

Notes

With this change, SELECT * EXCLUDE (...) no longer fails at generate time. It used to fail only because it produced an unnamed column. It now generates []any and fails at run time (expected 3 destination arguments in Scan, not 1). That is a separate problem with its own cause (convertStar returns a TODO node), described in #4660.