Network buffers associated with IO context leak in case of network failure

When network link is broken (pf or iptables), libcouchbase will continue retying operations until timeout, but the buffers that has been allocated during operation retry are not getting properly deallocated when the server purges pending operations.

Libcouchbase should clean the IO context object properly in case of network error.

To reproduce, just run some workload in the loop and cut the network link using firewall.

Useful commands for pf on macOS

Block traffic:

$ cat ~/block.conf

block drop in from 192.168.107.128

block drop in from 192.168.107.129

block drop in from 192.168.107.130

sudo pfctl -f ~/block.conf && sudo pfctl -e

Restore traffic:

sudo pfctl -d && sudo pfctl -F all