Fix dtype incorrectly being reset

[SkafteNicki]

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?

Fixes #187
Currently, if a module is casted to another dtype, say double

c = ConfusionMatrix(num_classes=3).double()

the initial metric states are correctly moved to double dtype. However, since currently is not applied to the _defaults (which contains the values we reset metric states to) after calling reset once the states will be back to whatever dtype they where original initialized with in self.add_state.
This PR fixes this and a bug in relation to this inside ConfusionMatrix that always forces the input to be float (even if asking for double).

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 #243 (47c6fa3) into master (a21eec5) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #243   +/-   ##
=======================================
  Coverage   96.82%   96.82%           
=======================================
  Files         184      184           
  Lines        6014     6024   +10     
=======================================
+ Hits         5823     5833   +10     
  Misses        191      191           

Flag	Coverage Δ
Linux	78.95% <14.28%> (-0.14%)	⬇️
Windows	78.95% <14.28%> (-0.14%)	⬇️
cpu	96.81% <100.00%> (+<0.01%)	⬆️
gpu	96.84% <ø> (+<0.01%)	⬆️
macOS	96.81% <100.00%> (+<0.01%)	⬆️
pytest	96.82% <100.00%> (+<0.01%)	⬆️
python3.6	95.75% <100.00%> (+<0.01%)	⬆️
python3.8	96.81% <100.00%> (+<0.01%)	⬆️
python3.9	96.71% <100.00%> (+<0.01%)	⬆️
torch1.3.1	95.75% <100.00%> (+<0.01%)	⬆️
torch1.4.0	95.88% <100.00%> (+<0.01%)	⬆️
torch1.8.1	96.71% <100.00%> (+<0.01%)	⬆️

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

Impacted Files	Coverage Δ
...chmetrics/functional/classification/cohen_kappa.py	100.00% <100.00%> (ø)
...rics/functional/classification/confusion_matrix.py	100.00% <100.00%> (ø)
torchmetrics/metric.py	95.56% <100.00%> (+0.06%)	⬆️
...chmetrics/functional/classification/specificity.py
...cs/functional/regression/mean_squared_log_error.py
__w/1/s/torchmetrics/classification/specificity.py
...1/s/torchmetrics/retrieval/mean_reciprocal_rank.py
...rchmetrics/functional/retrieval/reciprocal_rank.py
...trics/functional/regression/mean_relative_error.py
__w/1/s/torchmetrics/classification/cohen_kappa.py
... and 177 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 a21eec5...47c6fa3. Read the comment docs.