[🐛 BUG] Fix bugs when collecting results from mp.spawn in multi-GPU training

[ChenglongMa]

Bug description

We are unable to collect results directly from mp.spawn, e.g.,

RecBole/run_recbole_group.py

Lines 25 to 38 in 96eb311

	res = mp.spawn(
	run_recboles,
	args=(
	args.model,
	args.dataset,
	config_file_list,
	args.ip,
	args.port,
	args.world_size,
	args.nproc,
	args.group_offset,
	),
	nprocs=args.nproc,
	)

torch.multiprocessing.Queue can be applied for this purpose.

Changelog in this PR

1. Fixd result collection bug when using mp.spawn;
2. Updated documents of distributed training;
3. Fix bugs in significance_test.py