feat(grammar): RAISE, ASSERT, EXCEPTION handlers and GET DIAGNOSTICS

[sebsnyk]

Stacked PR, converted to draft. This sits second in the order #371 -> #372 -> #373 -> #374 and depends on #371. GitHub cannot base a fork PR on another fork branch, so the diff below includes #371; only the last commit (d621f5e) is this change. It is not ready for review until #371 lands and it is rebased; closing it until then is a reasonable call.

Problem

Once a body can hold control flow (#371), the next thing every real PL/pgSQL or PL/SQL routine trips on is error handling: RAISE NOTICE, RAISE EXCEPTION ... USING, an EXCEPTION WHEN others THEN section, and in tests ASSERT and GET DIAGNOSTICS. The #236 sample stops at its RAISE EXCEPTION. Against real sources:

source	ERROR nodes with #371	with this PR
PostgreSQL src/test/regress/sql/plpgsql.sql	636	414
utPLSQL source/core/ut_utils.pkb	233	233
pg_partman sql/functions/check_default.sql	11	11

utPLSQL and pg_partman do not move because their remaining errors are %TYPE, cursors, PERFORM/EXECUTE and Oracle's IS ... END name; headers, which come next.

Solution

A new grammar/statements/error-handling.js:

• raise_statement: RAISE [DEBUG | LOG | INFO | NOTICE | WARNING | EXCEPTION] [format, args... | condition | SQLSTATE 'x'] [USING option = expr, ...], and a bare RAISE. Each USING option is a raise_option.
• assert_statement: ASSERT condition [, message].
• exception_handler: WHEN cond [OR cond]... THEN statements, where a condition is OTHERS, SQLSTATE 'x' or a name. A keyword_exception followed by one or more handlers may close function_body, procedure_body and block.
• get_diagnostics: GET [CURRENT | STACKED] DIAGNOSTICS var = item, ....

All four are reachable only from bodies and blocks, as in #371.

One shape to check: a condition name in RAISE unique_violation parses as (field (identifier)), because a separate identifier alternative would be ambiguous with a format expression. In an exception_handler the name is a plain (identifier).

Notes

• tree-sitter test: 524 corpus tests on feat(grammar): procedural control flow in function and procedure bodies #371, 531 with this PR (7 new in test/corpus/error_handling.txt), plus the 2 highlight tests. No ERROR/MISSING; scripts/test-keywords.sh passes.
• Parse table: STATE_COUNT 30898 → 31012 (+0.4% on top of feat(grammar): procedural control flow in function and procedure bodies #371; +1.3% over main), parser.c 42.3 MB → 42.5 MB.
• With this the Syntax highlighting appears to give up in Postgres function bodies #236 sample has one unparsed construct left, RETURNING * INTO session_record, which belongs with SELECT ... INTO STRICT and EXECUTE ... INTO in a following PR.