Add duckdb_quote_identifier() to fix identifier quoting for DuckDB pushdown

DuckDB reserves several keywords that PostgreSQL does not (LAMBDA, PIVOT,
PIVOT_LONGER, PIVOT_WIDER, QUALIFY, SUMMARIZE, DESCRIBE, SHOW, UNPIVOT).
Using PostgreSQL's quote_identifier() on column names sent to pgduck_server
silently omits the extra quoting those reserved words require, causing query
failures at runtime.

This commit:

- Adds tools/generate_duckdb_kwlist.py, a script that parses the vendored
  DuckDB kwlist.hpp and emits pg_lake_engine/src/pgduck/duckdb_kwlist.h — a
  sorted C include fragment used to initialise the keyword lookup table.
  The generated file is checked in; re-run `make generate-duckdb-kwlist`
  after a DuckDB version bump.

- Adds `make generate-duckdb-kwlist` and `make check-duckdb-kwlist` targets.
  The check target verifies the generated file is up to date and can be wired
  into CI.

- Rewrites keywords.c to use the DuckDB-accurate keyword table instead of
  PostgreSQL's parser/kwlist.h.  Adds duckdb_quote_identifier(), which falls
  through to quote_identifier() for normal identifiers and additionally quotes
  identifiers that are RESERVED_KEYWORD in DuckDB.

- Exports duckdb_quote_identifier() from keywords.h.

- Replaces quote_identifier() with duckdb_quote_identifier() in all files
  that generate SQL sent to pgduck_server:
    - pg_lake_engine/src/pgduck/read_data.c  (23 call sites)
    - pg_lake_engine/src/pgduck/write_data.c  (5 call sites)
    - pg_lake_engine/src/pgduck/iceberg_query_validation.c  (4 call sites)
    - pg_lake_table/src/fdw/deparse.c  (7 call sites, plus audit TODO note)

  PostgreSQL-only callers (pg_extension_updater, base_worker_launcher,
  data_files_catalog, EXPLAIN output in pg_lake_table.c) are left unchanged.

Fixes #277.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: David Christensen <david.christensen@snowflake.com>

Author: sfc-gh-dachristensen

Makefile

@@ -14,7 +14,7 @@ CUSTOM_TARGETS = check-pg_lake_engine installcheck-pg_lake_engine check-pg_exten
 DUCKDB_BUILD_USE_CACHE ?= 0
 
 # other phony targets go here
-.PHONY: all fast install install-fast installcheck clean check submodules uninstall check-indent reindent installcheck-postgres installcheck-postgres-with_extensions_created
+.PHONY: all fast install install-fast installcheck clean check submodules uninstall check-indent reindent installcheck-postgres installcheck-postgres-with_extensions_created generate-duckdb-kwlist check-duckdb-kwlist
 .PHONY: $(ALL_TARGETS)
 .PHONY: $(PHONY_TARGETS)
 
@@ -220,6 +220,17 @@ uninstall-avro:
 	rm -f $(PG_LIBDIR)/libavro.*
 	rm -rf $(PG_INCLUDEDIR)/avro*
 
+## DuckDB keyword list maintenance
+# Regenerate the checked-in keyword table from the vendored DuckDB kwlist.hpp.
+# Re-run whenever duckdb_pglake/duckdb is updated to a new DuckDB release.
+generate-duckdb-kwlist:
+	python3 tools/generate_duckdb_kwlist.py
+
+# Verify that the checked-in keyword table matches the current kwlist.hpp.
+# Run in CI to catch stale keyword tables after a DuckDB version bump.
+check-duckdb-kwlist:
+	python3 tools/generate_duckdb_kwlist.py --check
+
 ## Other targets
 check-isolation_pg_lake_table:
 	$(MAKE) -C pg_lake_table check-isolation

pg_lake_engine/include/pg_lake/pgduck/keywords.h

@@ -17,4 +17,19 @@
 
 #pragma once
 
+/*
+ * IsDuckDBReservedWord — returns true for any keyword that is not
+ * UNRESERVED_KEYWORD in DuckDB (i.e., RESERVED, COL_NAME, or
+ * TYPE_FUNC_NAME).  Used for struct field-access quoting.
+ */
 PGDLLEXPORT bool IsDuckDBReservedWord(char *candidateWord);
+
+/*
+ * duckdb_quote_identifier — like quote_identifier() but also quotes
+ * identifiers that are RESERVED_KEYWORD in DuckDB but not in PostgreSQL
+ * (e.g. LAMBDA, PIVOT, QUALIFY, SUMMARIZE, DESCRIBE, SHOW, UNPIVOT).
+ *
+ * Use this for all identifiers (column names, field names, relation names)
+ * that will appear in SQL sent to pgduck_server.
+ */
+PGDLLEXPORT const char *duckdb_quote_identifier(const char *ident);