[ENH] t-SNE: Add Normalize data checkbox

[pavlin-policar]

Issue

Fixes #3448

Description of changes

• The widget was still using SVD on sparse data. I have changed this to PCA, since our PCA does support sparse data since [ENH] Implement better randomized PCA #3532

• I basically copied over the functionality from the PCA widget. PCA currently doesn't support normalization on sparse data, so this option is disabled on sparse data, just like in the PCA widget.

• Lastly, I increased the error margin on one test. The test does this: it embeds a data set using t-SNE then embeds the same data onto the existing embedding (transform). Unfortunately, the points are bound to be jittered around each original corresponding point, but also how far away is also determined by the neighborhoods, so there's no clean way to check for this. But if we visualize what this actually produces, we can see that the result is still correct. And given that the space spans from -20 to 20, increasing the error margin from 1 to 3 shouldn't really impact results too much.
  

Includes

• Code changes
• Tests
• Documentation

[pavlin-policar]

I removed this check because I changed line 146 to invalidate the PCA projection, therefore it will be set to None whenever the number of components changes.

[codecov]

Codecov Report

Merging #3570 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #3570      +/-   ##
==========================================
+ Coverage   83.97%   83.98%   +<.01%     
==========================================
  Files         370      370              
  Lines       66941    66973      +32     
==========================================
+ Hits        56215    56244      +29     
- Misses      10726    10729       +3

[lanzagar]

I would prefer to add normalization support in a separate PR, adding it everywhere where it's missing, and leave this as is.

Fine with me.