Fix build warnings against PG 16+

[devrimgunduz]

PG_ENSURE_ERROR_CLEANUP()/PG_END_ENSURE_ERROR_CLEANUP(), as defined in storage/ipc.h, expand to a plain PG_TRY()/PG_CATCH()/PG_END_TRY() with no suffix argument. Nesting two such blocks in the same function (as happens below, in pglogical_sync_subscription()) makes the inner block's PG_TRY()-generated locals shadow the outer block's, which trips -Wshadow=compatible-local on modern gcc. This is purely cosmetic: each PG_TRY() only ever touches its own copy of the saved exception/context stack, so the shadowing has no runtime effect.

As of PG 16, core's PG_TRY()/PG_CATCH()/PG_END_TRY() accept an optional suffix argument specifically to let nested blocks use distinct variable names and silence this warning -- but PG_ENSURE_ERROR_CLEANUP() itself doesn't forward that suffix through. On PG 16+ we define local wrappers that do. On PG 14/15, PG_TRY() and friends take no arguments at all, so there's no way to avoid the warning short of restructuring the code (e.g. moving the inner block into a separate function); we just fall back to the plain core macros there and accept the harmless warning.

These wrappers can be removed if/when upstream teaches PG_ENSURE_ERROR_CLEANUP() to forward a suffix through itself on all still-supported branches.

Hacked by Claude, tested by me.

[eulerto]

I submitted a patch to -hackers.

https://www.postgresql.org/message-id/a690b47b-1859-4598-8541-286e48e7a724%40app.fastmail.com