Audit `kill <pid>` SQL

There are reports of `kill` not actually killing a query from Gas Town users. I have also experienced a `kill` on a large, memory-consuming query not working.

We need a suite of tests on `kill`.

We need to fix any bugs that prevent killing a rogue query in a timely manner.

LLMs write gnarly SQL and often `kill` will be the last line of defense. It needs to work.

Start by locating the SQL implementation and existing tests for `kill `, then reproduce the reported failure with a large, memory-consuming query. Add coverage for killing rogue queries and verify that the command terminates them in a timely way; investigate and fix any failures revealed by the tests.

Written by the indexing model from the issue text.