Add unit tests for backend functions

[nkoep]

The backend functions are currently not tested individually in isolation. This means that bugs in the backend abstractions are only spotted by unit tests if some of the more involved tests actually use a particular backend function. This would also resolve https://github.com/geomstats/geomstats/issues/303.

[opeltre]

[torch] : gs.expm does not seem properly vectorised, neither gs.matmul

[opeltre]

expected:

gs.matmul(a, gs.array([b, c]))
>>> [ab, ac]

[opeltre]

simple vectorisation expm(gs.array([a, b, c])) fails on pythorch

[opeltre]

[numpy] gs.vectorize does not return a function, as np.vectorize does

#numpy:
f = lambda x : x + 1
map_f = np.vectorise(f)
map_f(a, b, c)
>>> [f(a), f(b), f(c)]

#geomstats:
map_f = gs.vectorize(f)
>>> Error: missing  positional  argument pyfunc
gs.vectorize([a, b, c], f)
>>> [f(a), f(b), f(c)]

[opeltre]

[torch] gs.vectorize not implemented

[opeltre]

Maybe not the place to post backend issues as they do not solve the backend testing problem, but these do bump with nose2 when running unrelated tests :)

[opeltre]

[pytorch]

1. still no gs.linalg.logm

2. plain gs.exp does not accept scalars

   >>> gs.exp(2.)
   TypeError: exp(): argument 'input' (position 1) must be Tensor, not float

3. gs.linalg.inv is not vectorized

   >>> a = gs.eye(3)
   >>> gs.linalg.inv(gs.tensor([a, 2*a]))
   raise LinAlgError('Last 2 dimensions of the array must be square') 
   numpy.linalg.LinAlgError: Last 2 dimensions of the array must be square

   Good to list what manual testing yields before someone has the courage to settle for the task :)

Due to the amount of problems the issue could probably be renamed "fix & test backend"...

[nguigs]

This has been partially done manually. Could-it be automated @SaitejaUtpala ?