[feat] Add reset_forward_cache to Metric

[tchaton]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

What does this PR do?

This PR adds reset_forward_cache option to Metric.

When wrapping Metric in Metric, the forward_cache get overridden. This option enables to get skip this.

This would be used in Logging Refactor in Lightning: Lightning-AI/pytorch-lightning#7631

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #260 (ca20eac) into master (571ac15) will decrease coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #260      +/-   ##
==========================================
- Coverage   96.83%   96.81%   -0.02%     
==========================================
  Files         184       92      -92     
  Lines        6026     3013    -3013     
==========================================
- Hits         5835     2917    -2918     
+ Misses        191       96      -95     

Flag	Coverage Δ
Linux	78.95% <100.00%> (ø)
Windows	78.95% <100.00%> (ø)
cpu	96.81% <100.00%> (ø)
gpu	?
macOS	96.81% <100.00%> (ø)
pytest	96.81% <100.00%> (-0.02%)	⬇️
python3.6	95.75% <100.00%> (ø)
python3.8	96.81% <100.00%> (ø)
python3.9	96.71% <100.00%> (ø)
torch1.3.1	95.75% <100.00%> (ø)
torch1.4.0	95.88% <100.00%> (ø)
torch1.8.1	96.71% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
torchmetrics/metric.py	95.56% <100.00%> (ø)
.../1/s/torchmetrics/regression/explained_variance.py
.../1/s/torchmetrics/functional/classification/roc.py
...s/torchmetrics/classification/matthews_corrcoef.py
__w/1/s/torchmetrics/classification/roc.py
__w/1/s/torchmetrics/utilities/distributed.py
...torchmetrics/functional/classification/__init__.py
...rchmetrics/functional/retrieval/reciprocal_rank.py
...1/s/torchmetrics/functional/classification/dice.py
__w/1/s/torchmetrics/functional/nlp.py
... and 83 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 571ac15...ca20eac. Read the comment docs.

[ananthsub]

@tchaton from a metrics standpoint, forward_cache is an internal implementation detail. Is there any way we can avoid this dependency on forward cache in the Lightning trainer?

[justusschock]

I also think this shouldn't be part of this package but part of lightning since torchmetrics is supposed to be lightning agnostic