Fix the PostgreSQL escape-string bypass in read-only SQL in the next release #630

The finding in #612 (sensitive columns pass the read-only SQL guard through a PostgreSQL E-string and a result alias) is not part of 1.5.2. The fix needs a provider-aware parse, or an explicit rejection of syntax the guard does not understand, plus a regression test. That is too much change for a release that is already built and in the lab.

It goes into the release after 1.5.2. Problem, evidence and acceptance criteria stay in #612.