Add optional DDL to SQL pipeline step #2460

The Oracle profiler PR (#2187) introduced a pattern where every logical extractor is split into two YAML steps that share the same name: — one with type: ddl, one with type: sql. This works today because pipeline.py dispatches on step.type first, but it causes three problems:

1. Ambiguous failure reporting. PipelineClass.execute() aggregates StepExecutionResult by step.name, so when the SQL half of config_containers fails, users see Pipeline execution failed due to errors in steps: config_containers with no signal whether DDL or SQL was the culprit.
2. Implicit name=table contract. _execute_ddl_step checks if not self._table_exists(conn, step.name): conn.execute(ddl), so the DDL must CREATE TABLE whose name is exactly step.name. Nothing enforces this; an edit that changes the name in one place but not the other silently no-ops or fails at insert time.
3. Future fragility. Anyone refactoring pipeline.py to dedupe steps by name (perfectly reasonable assumption) breaks the YAML silently.

Scope / acceptance criteria

• Fold DDL into the SQL step.** Single step with setup_ddl: and extract_sql: fields. Larger refactor, but conceptually cleaner since DDL+SQL are one extractor.
• Migrate the Oracle pipeline_config (PR feature/oracle_profiler #2187).
• Migrate the MSSQL pipeline_config once it converts to type: sql (see Convert MSSQL profiler from type:python venv steps to type:sql in-process steps #2459).
• Document the decision so future profilers follow the same shape.

References

• PR feature/oracle_profiler #2187 review thread on oracle/pipeline_config.yml:11.
• Related: Convert MSSQL profiler from type:python venv steps to type:sql in-process steps #2459 (Convert MSSQL profiler from type: python to type: sql).