Fix Ensemble-Stat logic for computing the forecast lead time of a time-lagged ensemble.

[JohnHalleyGotway]

Jeff Hamilton from NOAA-GSD has been running Ensemble-Stat to generate time-lagged ensemble output. As of met-9.0_beta2, Ensemble-Stat writes to the output the timing information of the last ensemble member processed. Instead, Ensemble-Stat should write the minimum forecast lead time.

When processing a time-lagged ensemble consisting of members with 6 and 18 hour lead times, the lead time for the output should be reported as 6 hours, not 18, even if the last ensemble member was 18.

[JohnHalleyGotway]

I ran a test case to define a time-lagged ensemble for 2 members: a 0-hour lead time and a 12-hour lead time, both valid at the same time. I found that the timing information reported in the output NetCDF file is the init and valid times of the FIRST ensemble member read.

For time-lagged ensembles, the valid time should remain constant, but the init time changes. We should change Ensemble-Stat to report the MAXIMUM init time encountered... and that would result in the minimum lead time.