Fix problem in selection of the principal directions in Tangent PCA.

[YannCabanes]

This PR is related to PR #1513 in which a solution is proposed.

The current Tangent PCA algorithm does not necessarily select the axis which corresponds to the largest variance.
The problem is that the current Tangent PCA does not take into account the metric at the mean to sort the principal directions.

To show this problem, I have created an almost Euclidean manifold (e488adb).
This manifold is the Euclidean space of dimension 2 whose metric has been modified to give more weight to the second axis. It's infinitesimal metric element is defined by: ds^2 = dx1^2 + 10000 dx2^2
Then if we run the following codes (e488adb):

almost_euclidean_dim_two = AlmostEuclideanDimTwo()
tpca = TangentPCA(space=almost_euclidean_dim_two, n_components=1)
point_a = gs.array([10, 0], dtype=float)
point_b = gs.array([0, 1], dtype=float)
X = gs.stack([point_a, -point_a, point_b, -point_b], axis=0)
mean = gs.array([0, 0], dtype=float)
tpca.fit(X=X, base_point=mean)
variance_axis_1 = almost_euclidean_dim_two.metric.squared_dist(mean, point_a)  # variance_axis_1 = 100
variance_axis_2 = almost_euclidean_dim_two.metric.squared_dist(mean, point_b)  # variance_axis_1 = 10000
assert variance_axis_2 >= variance_axis_1 # True
axis_1 = gs.array([1, 0])
axis_2 = gs.array([0, 1])
assert gs.all(tpca.components_ == axis_1) # True

The tangent PCA returns axis_2 as the principal axis since it is the axis which
corresponds to the larger variance. It is returning axis_1 instead.

To solve this problem, we can multiply the log vectors at the mean by square root of the metric_matrix to select the correct principal directions (see PR #1513).

[YannCabanes]

Hello @luisfpereira and @ninamiolane,
Do you agree that the current behavior of the tangent PCA described above is not the expected one?
(On the example in dimension 2, it does not rank first the direction with the largest variance.)

[codecov]

Codecov Report

Merging #1878 (12b3266) into master (6386ade) will increase coverage by 9.00%.
The diff coverage is 93.34%.

@@            Coverage Diff             @@
##           master    #1878      +/-   ##
==========================================
+ Coverage   82.59%   91.58%   +9.00%     
==========================================
  Files         136      141       +5     
  Lines       13577    13645      +68     
==========================================
+ Hits        11213    12496    +1283     
+ Misses       2364     1149    -1215     

Flag	Coverage Δ
numpy	87.38% <93.34%> (?)
pytorch	82.47% <4.45%> (-0.11%)	⬇️

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

Files Changed	Coverage Δ
geomstats/learning/pca.py	94.00% <93.34%> (+77.60%)	⬆️

... and 45 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[YannCabanes]

The tests pass except Linting, Testing / build (ubuntu-latest, 3.11, autograd (errors not related to this PR) and DeepSource.