Clear error state before logging prewarm cancellation

[mingjiegao]

Problem

In pgxn/neon/file_cache.c, the prewarm cancellation path can enter PG_CATCH() when WaitForBackgroundWorkerShutdown() is interrupted, for example because of cancellation or another error.

The catch block currently calls elog(LOG, ...) before clearing the active error state. PostgreSQL keeps the backend in the error handling context until FlushErrorState() is called. If this path is hit repeatedly, the error data stack can keep growing. After 5 entries (ERRORDATA_STACK_SIZE = 5), PostgreSQL can raise:

ereport(PANIC, (errmsg_internal("ERRORDATA_STACK_SIZE exceeded")));

Summary of changes

Call FlushErrorState() before elog(LOG, "LFC: cancel prewarm") in the PG_CATCH() block, so the previous error state is cleared before the catch path calls back into PostgreSQL's error reporting machinery.