Avoid expensive deepcopy when rotating Miller

[viljarjf]

Description of the change

Rotating a Miller instance with a Rotation calls Miller.__getitem__ to handle improper rotations, regardless of whether any rotations are actually improper.
As Miller.__getitem__ involves a deepcopy, avoiding this when not necessary provides a significant speedup, depending on the complexity of the associated Phase.

Progress of the PR

• [n/a] Docstrings for all functions
• Unit tests with pytest for all lines
• Clean code style by running black via pre-commit

Minimal example of the bug fix or new feature

from orix.crystal_map import Phase
from orix.vector import Miller
from orix.quaternion import Rotation

P1 = Phase("P1", 1)
Fm3m = Phase("Fm3m", 225)
P1_vec = Miller.random(P1)
Fm3m_vec = Miller.random(Fm3m)
rot = Rotation.random(2000)


%timeit rot * P1_vec
%timeit rot * Fm3m_vec

Current
505 μs ± 18.6 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
1.7 ms ± 48.2 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

New change
346 μs ± 41.4 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
350 μs ± 16.9 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

Of course, with improper rotations, the runtime is the same:

from orix.crystal_map import Phase
from orix.vector import Miller
from orix.quaternion import Rotation

P1 = Phase("P1", 1)
Fm3m = Phase("Fm3m", 225)
P1_vec = Miller.random(P1)
Fm3m_vec = Miller.random(Fm3m)
rot = Rotation.random(2000) * -1

assert all(rot.improper)

%timeit rot * P1_vec
%timeit rot * Fm3m_vec

506 μs ± 26.3 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
1.69 ms ± 44.3 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

For reviewers

• The PR title is short, concise, and will make sense 1 year later.
• New functions are imported in corresponding __init__.py.
• New features, API changes, and deprecations are mentioned in the unreleased
  section in CHANGELOG.rst.
• Contributor(s) are listed correctly in __credits__ in orix/__init__.py and in
  .zenodo.json.

[hakonanes]

Seems simple enough. I'll have a look within tomorrow.

[hakonanes]

Looking beyond this PR, I think it may be better to avoid deepcopy() altogether and rather add a copy() method, like this

def copy(self) -> Self:
    return self.__class__(self)

We can update the creation method to allow another instance as the first parameter, and only copy the things we need.

We can deprecate deepcopy() in favor of copy().

Benefits:

• More control
• New copy via Vector3d(Vector3d) or Vector3d.copy()

What do you think, @viljarjf?

[hakonanes]

Let's continue the copy discussion in #535.