Datasets: improve lazy import error msg for missing deps

[adamjstewart]

TorchGeo has a number of optional dependencies only needed by certain datasets. We use lazy imports to allow these dependencies to be optional. This PR improves these error messages to make it more clear how to solve the issue.

Before

ImportError: scipy is not installed and is required to use this dataset

After

torchgeo.datasets.errors.DependencyNotFoundError: scipy is not installed and is required to use this dataset. Either run:

$ pip install scipy

to install just this dependency, or:

$ pip install torchgeo[datasets]

to install all optional dataset dependencies.

This has a number of other benefits:

• Reduces code duplication (making it easier to update the message for all imports if we want to, e.g., add conda installation instructions)
• Allows us to easily fail fast during dataset instantiation
• Makes it easier to get 100% coverage on all datasets in CI: test optional datasets on every commit #2045

Closes #2045
Prerequisite for #2043

[adamjstewart]

TIL about raise ... from ...

Without

> python3 -c 'from torchgeo.datasets.utils import lazy_import; lazy_import("foo")'
Traceback (most recent call last):
  File "/Users/Adam/torchgeo/torchgeo/datasets/utils.py", line 781, in lazy_import
    return importlib.import_module(name)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/Adam/spack/var/spack/environments/default/.spack-env/view/lib/python3.11/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1204, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1140, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'foo'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/Users/Adam/torchgeo/torchgeo/datasets/utils.py", line 788, in lazy_import
    raise MissingDependencyError(name)
torchgeo.datasets.errors.MissingDependencyError: foo is not installed and is required to use this dataset. Either run:

$ pip install foo

to install just this dependency, or:

$ pip install torchgeo[datasets]

to install all optional dataset dependencies.

With

> python3 -c 'from torchgeo.datasets.utils import lazy_import; lazy_import("foo")'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/Users/Adam/torchgeo/torchgeo/datasets/utils.py", line 788, in lazy_import
    raise MissingDependencyError(name) from None
torchgeo.datasets.errors.MissingDependencyError: foo is not installed and is required to use this dataset. Either run:

$ pip install foo

to install just this dependency, or:

$ pip install torchgeo[datasets]

to install all optional dataset dependencies.

[adamjstewart]

Only downside of the lazy import function is that we lose type hints.

[adamjstewart]

Apparently ImportError and ModuleNotFoundError are different:

>>> import foo
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'foo'
>>> import numpy.foo
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ModuleNotFoundError: No module named 'numpy.foo'
>>> from numpy import foo
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ImportError: cannot import name 'foo' from 'numpy' (/Users/Adam/spack/var/spack/environments/default/.spack-env/view/lib/python3.11/site-packages/numpy/__init__.py)

[adamjstewart]

We could also check the version, but that would require an extra dep on packaging.

[calebrob6]

This is just plain nice