[RFC] Depreceate `Trainer(benchmark)` and use deterministic

[tchaton]

Out of scope for this PR but maybe we should remove Trainer(benchmark) and set

torch.backends.cudnn.benchmark = not deterministic

Originally posted by @carmocca in #9121 (comment)

cc @Borda @justusschock @awaelchli @akihironitta @rohitgr7 @tchaton

[rishawsingh]

Hi I would like to work on this issue, if it's still open

[rishawsingh]

@carmocca can you help me out with this issue, since am new to open source contribution

[carmocca]

Sure. You can either ask here or through the Slack channel!

https://join.slack.com/t/pytorch-lightning/shared_invite/zt-pw5v393p-qRaDgEk24~EjiZNBpSQFgQ
(if you need access)

[kaushikb11]

Hi @rishawsingh, any update on this? Let us know if we could help in any way.