make sure to configure the channel to nonblocking before we close. A close to a blocking channel can block and can therefore block all spools and therefore potentially the full server.
Background: Tcl's fcopy sets a channel to non-blocking. In earlier Tcl versions, the channel was left in the non-blocking state, also after fcopy terminates (e.g. in error cases).
Newer Tcl versions seem to reestablish the earlier state, so we have to care now for the problem in bgdelivery.
