Speed up recursions

[BalzaniEdoardo]

In this PR:

• New accept integer for __mul__: basis * 3 will be equivalent to basis + basis+basis
• Add an __rmul__ enabling right multiplicationn * basis
• Improved the efficiency of __pow__ by bisect multiply (log2(n) multiplication, less deep copy)
• Add a mechanism that do not perform a deep-copy when cloning a bases

Old clone performance

bas = nmo.basis.BSplineEval(5)**500
%timeit bas.__sklearn_clone__()
2.47 s ± 107 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

New clone performance

bas = nmo.basis.BSplineEval(5)**500
%timeit bas.__sklearn_clone__()
10.9 ms ± 22 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Old Power Performance

bas = nmo.basis.BSplineEval(5)
%timeit bas ** 500
2.49 s ± 76.7 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

New Power Performance

bas = nmo.basis.BSplineEval(5)
%timeit bas ** 500
8.46 ms ± 79.3 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 96.82540% with 2 lines in your changes missing coverage. Please review.

Project coverage is 97.00%. Comparing base (a510ef3) to head (54f9c97).
Report is 361 commits behind head on development.

Files with missing lines	Patch %	Lines
src/nemos/basis/_basis.py	95.34%	2 Missing ⚠️

Additional details and impacted files

@@               Coverage Diff               @@
##           development     #307      +/-   ##
===============================================
+ Coverage        96.13%   97.00%   +0.86%     
===============================================
  Files               34       36       +2     
  Lines             2642     3208     +566     
===============================================
+ Hits              2540     3112     +572     
+ Misses             102       96       -6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[billbrod]

Generally looks good! Just a couple of small changes:

• Need to update labels at the end of the mul/pow recursion (and add test).

bas = nmo.basis.BSplineEval(5)
add = bas*3
add.basis1.basis1.label == add.basis2.label

• pow should have the same check about label as mul
• I think it's worth changing the references to _shallow_copy and _set_shallow_copy_temporarily to self.__class__ or similar, to make it clear that it's class-level behavior we're changing.
• I also think you can remove the temporarily part of that method: it's a context manager, so temporarily is implied.

[billbrod]

While reviewing this PR, I looked into how to use not-quite-global variables, like the new shallow_copy flag. That flag affects the behavior of some functions of all composite classes, so it can't be instance level (i.e., self._shallow_copy). Right now, it's class level (self.__class__._shallow_copy), which I think is right. While looking into this, came across this thread, which is discussing how stdlib's warnings handles filters, and they're set at the module-level, using sys.modules["warnings"].filters. So if we had flags that affected the behavior across the module (not just within a single class/mixin), then that would be the way to go.

[BalzaniEdoardo]

I think I should have addressed everything, including adding missing tests!
@billbrod note that I added a call to the function that recomputes all default labels after the shallow copy composition is completed, and I realized tan extra corner case that I fixed

add = nmo.basis.BSplineEval(5)*2

# this would have called deep-copy in original implementation but it would copy the parent as well.
# now it prints nothing
out = add.basis1 * 1

# originally would have printed the same as print(add)
print(out._parent)