Remove calls to profile `model_forward`

[daniellepintz]

Proposed refactor

Remove these two calls to profile model_forward:
https://github.com/PyTorchLightning/pytorch-lightning/blob/d902b5988601a21db04ffe6316a012581fe152ec/pytorch_lightning/loops/optimization/optimizer_loop.py#L417
https://github.com/PyTorchLightning/pytorch-lightning/blob/d902b5988601a21db04ffe6316a012581fe152ec/pytorch_lightning/loops/optimization/manual_loop.py#L110

Motivation

We profile training_step a few lines after we profile model_forward:
https://github.com/PyTorchLightning/pytorch-lightning/blob/d902b5988601a21db04ffe6316a012581fe152ec/pytorch_lightning/loops/optimization/optimizer_loop.py#L417-L424

https://github.com/PyTorchLightning/pytorch-lightning/blob/d902b5988601a21db04ffe6316a012581fe152ec/pytorch_lightning/loops/optimization/manual_loop.py#L110-L117

There is a lot of overlap between model_forward and training_step, with training_step taking up the majority of model_forward's time.

see lines 11-12 in this example profiler output: https://gist.github.com/daniellepintz/7b37ab4807fab345248bf60c72e46ae8

Additional Context

Discussed in #12015

cc @carmocca @kaushikb11 @ninginthecloud @rohitgr7

[carmocca]

Nice. These have been carried over from refactor to refactor since a long time ago.

It's good to finally remove them. Less framework overhead.

[daniellepintz]

Cool! Let me know if you see any more optimizations we can do in #12015