Add the possibility to give an array-like input for n_comp option in KPLS (#816)

… and test the range of n_dim and choose the best one. It has been
refactored to take into account the advice given by @Paul-Saves.

I added unit tests (some of them are quite long due to the exhaustive
search, so I marked them as
`@unittest.skipIf(int(os.getenv("RUN_SLOW_TESTS", 0)) < 1, "too
slow")`).

I updated the docs with a new example and added some description to the
parameters.

I also added a little safeguard on the number of k_fold min based on the
number of training points:
```python
k_fold = min(4, original_nt)
```

---------

Co-authored-by: Saves Paul <paul.saves@alumni.enac.fr>

Author: enricostragiotti

doc/_src_docs/surrogate_models/gpr/kpls.rst

@@ -31,6 +31,11 @@ Usage with an automatic number of components
 
 .. embed-test-print :: smt.surrogate_models.tests.test_surrogate_model_examples , Test , test_kpls_auto
 
+Usage with a range of number of components
+--------------------------------------------
+
+.. embed-test-print :: smt.surrogate_models.tests.test_surrogate_model_examples , Test , test_kpls_range
+
 Options
 -------