Progress_watermark needed to be reset when next iterator start

[linbing]

in nova/virt/libvrit/driver of _live_migration_monitor, the progress_watermark needed to be reset to 0 when progress_watermark is smaller than info.data_remaning.

when migration iteration began, the progress_watermark maybe the mark of pre data_remaining of one iteration, so when in a big ram writing env, it is may cause multiple migration iteration, so it is may cause data_remaining > progress_watermark, there is a log output to recode this, so when this happened, and next iteration come in, the progress_watermark is small than data_remaining forever, and progress_time may never be reset, it's will cause progress_timeout, however, it's not progress_timeout.