PT: also use complete_frac for progress reporting

[NeoLegends]

After this PR, the torch engine will use complete_frac for progress reporting as well as num_seqs.

Example line:

ep 1 train, step 23, ctc_8 10.797, ctc_16 10.016, ce 8.758, fer 0.979, grad_norm:p2 85.874, num_seqs 34, max_size:time 239629, max_size:out-spatial 80, mem_usage:cuda:0 60.0GB, 0.926 sec/step, elapsed 0:00:55, exp. remaining 4:59:46, complete 0.31%