Fix random point / vector allocations

[kellertuer]

This is the companion PR to JuliaManifolds/ManifoldsBase.jl#168 to fix using TangentSpaceAtPoint on non-matrix-type manifolds.

Resolves #652

[codecov]

Codecov Report

Merging #653 (da47217) into master (b5cdb3c) will decrease coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #653      +/-   ##
==========================================
- Coverage   99.22%   99.18%   -0.04%     
==========================================
  Files         106      106              
  Lines       10433    10440       +7     
==========================================
+ Hits        10352    10355       +3     
- Misses         81       85       +4     

Files Changed	Coverage Δ
src/manifolds/FixedRankMatrices.jl	98.20% <100.00%> (-0.03%)	⬇️
src/manifolds/PositiveNumbers.jl	100.00% <100.00%> (ø)
src/manifolds/SymmetricPositiveDefinite.jl	100.00% <100.00%> (ø)
src/manifolds/VectorBundle.jl	100.00% <100.00%> (ø)

... and 1 file with indirect coverage changes

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

[kellertuer]

For positive numbers the problem was, that the allocating version has to return a number and can not allocate anything,
for SPDPoint we needed an allocation of a tangent vector – but then now rand should be overall more stable when it comes especially to random tangents.

[mateuszbaran]

LGTM. Maybe up to the point that we still might want to have in-place rand! for PositiveNumbers?

[kellertuer]

Sure, then we just have to check that it is also covered by tests, the new (allocating) one basically covers all cases we currently have in the tests.

edit: Had an idea, stole the zero-dimensional arrays from the circle and added both rand! and a test.

[kellertuer]

The new four uncovered lines are false positives (function headers but the functions are green