fix(sql): walk block nodes so DDL before a DO block is extracted - #3900

What does this PR do?

Fixes #3886.

If a migration opens with a bare BEGIN; and later has a DO $$ ... END $$; block, tree-sitter-sql treats everything from that BEGIN up to the END of the DO body as a single block node. The top-level loop in extract_sql handled statement and transaction but not block, so every CREATE TABLE inside it got skipped without any error. That's why the reporter's schema only showed the tables that came after the DO block.

Here's the tree for the reduced repro:

block ['keyword_begin', ';', 'statement', ';', 'ERROR', 'keyword_end']
ERROR []
;
ERROR ['dollar_quote', ';', 'keyword_commit', ';']

The fix adds a block branch that walks the node, the same way transaction is already handled. walk() already recurses and _add_node dedupes by id, so nothing new is needed past the dispatch.

Type of change

• Bug fix
• New feature
• Documentation
• Tests or CI
• Refactor
• Security fix

Verification & Invariants

Invariant: a CREATE TABLE that parses cleanly should always end up as a table node, whatever wrapper node the grammar happens to put around it.

• Read the CONTRIBUTING.md guide.
• Reproduced the issue and identified the invariant.
• Made the smallest fix necessary.
• Added a regression test (if bug fix) or isolated boundary test.
• Kept the PR description synchronized with the final implementation.
• Documented any limitations / unsupported cases explicitly.

Limitation: the CREATE TRIGGER inside the DO body still isn't extracted, since it lives in a dollar-quoted string the grammar doesn't parse. This PR only brings back the tables that were being dropped around it.

How was this tested?

uv sync --all-extras --frozen

# regression test, extractor reverted to v8 -> fails
uv run pytest tests/test_multilang.py::test_sql_create_table_before_do_block_in_transaction
1 failed

# with the fix -> passes
1 passed

uv run pytest tests -q -k sql
43 passed

uv run pytest tests -q
6088 passed, 11 failed

All 11 failures are in tests/test_skillgen.py and they fail the same way on plain v8. My local clone is shallow, so the audit can't read an older commit (could not read 47042beb…:graphify/skill.md). They're unrelated to this change.

Graphify-specific checklist

• I updated generated skill artifacts (uv run python -m tools.skillgen --bless) when changing their source fragments.
• I confirmed that AST/structural extraction remains deterministic (no ambient state dependencies like ENV variables).
• I reviewed changes for security implications (no unsafe interpolation into shell/Python).
• I confirmed no API keys or local-only graph data are included.
• (If applicable) I disclosed AI authorship in my commit messages.

Note: I used Claude Code while working on this fix. I reviewed and ran all of it myself.