fix: check for timeout in connection after last statement finished (#…

…1086)

The check whether the previous statement timed out in the Connection API was
done when a statement was submitted to the connection, and not when the statement
was executed. That could cause a race condition, as statements are executed using
a separate thread, while submitting a statement is done using the main thread.

This could cause a statement to return an error with code ABORTED instead of
FAILED_PRECONDITION. A statement on a read/write transaction would always return
an error when a/the previous statement timed out, only the error code could
be different depending on whether the race condition occurred or not. This is
now fixed so that the error code is always FAILED_PRECONDITION and the error
indicates that a read/write transaction is no longer usable after a statement
timeout.

Fixes #1077