Allow passing rdzv_conf in dist component

[clumsy]

Description

The default timeout for c10d rdzv is 60 seconds. The more nodes are used to run a job - the more unlikely it is to get them aligned in time for this time window. Luckily torchrun provide a way to control this: https://github.com/pytorch/pytorch/blob/main/torch/distributed/run.py#L443

Motivation/Background

Unfortunately there's no env variable that could replace this.

Detailed Proposal

Simply add a new option next to rdzv_backend and rdzv_port

Alternatives

Create and maintain a custom component.

Additional context/links