Fix pre-commit hook failures (#756)

Pre-commit hook fails when run on all files for three reasons:
(see trace below)

1. Trailing whitespace on multiple files
2. mypy fails to load numpy and then subsequently fails to load
LazyModule from pipe.py
3. isort sees issues with known_third_party packages

```
> pre-commit run --all-files

Trim Trailing Whitespace.................................................Failed
- hook id: trailing-whitespace
- exit code: 1
- files were modified by this hook

Fixing docs/source/conf.py
Fixing fairscale/experimental/nn/auto_shard.py
Fixing docs/source/deep_dive/activation_checkpointing.rst
Fixing docs/source/tutorials/pipe.rst
Fixing docs/source/installation_instructions.rst
Fixing docs/source/deep_dive/pipeline_parallelism.rst
Fixing docs/source/tutorials/activation_checkpointing.rst
Fixing docs/source/tutorials/offload_model.rst
Fixing docs/source/deep_dive/oss_sdp_fsdp.rst
Fixing docs/source/what_is_fairscale.rst
Fixing CHANGELOG.md
Fixing fairscale/experimental/nn/offload.py
Fixing docs/source/index.rst
Fixing docs/source/deep_dive/adascale.rst
Fixing README.md
Fixing docs/source/tutorials/oss.rst
Fixing docs/source/deep_dive/offload.rst

Check python ast.........................................................Passed
Check for merge conflicts................................................Passed
Don't commit to branch...................................................Passed
Check for added large files..............................................Passed
Fix End of Files.........................................................Failed
- hook id: end-of-file-fixer
- exit code: 1
- files were modified by this hook

Fixing requirements.txt
Fixing docs/source/getting_started.rst
Fixing docs/source/installation_instructions.rst
Fixing codecov.yml
Fixing docs/source/deep_dive/adascale.rst
Fixing docs/source/tutorials/oss.rst
Fixing docs/source/deep_dive/offload.rst

black....................................................................Passed
flake8...................................................................Passed
seed isort known_third_party.............................................Failed
- hook id: seed-isort-config
- exit code: 1
- files were modified by this hook
isort....................................................................Passed
mypy.....................................................................Failed
- hook id: mypy
- exit code: 2

setup.cfg:45: error: Error importing plugin 'numpy.typing.mypy_plugin': No module named 'numpy'
Found 1 error in 1 file (checked 197 source files)
```

.pre-commit-config.yaml

@@ -43,3 +43,4 @@ repos:
     rev: 'v0.790'
     hooks:
     -   id: mypy
+        additional_dependencies: [numpy]

docs/source/deep_dive/adascale.rst

@@ -42,4 +42,3 @@ The training process is as follows:
 Best practices for `fairscale.optim.AdaScale`
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 Adascale only works for the SGD optimizer (with and without momentum)
-

docs/source/deep_dive/offload.rst

@@ -42,4 +42,3 @@ Best practices for using `fairscale.experimental.nn.OffloadModel`
 1. Using OffloadModel to train large models can result in loss of throughput which can be overcome by using activation checkpointing and microbatches.
 
 2. OffloadModel currently only works for `nn.Sequential` models.
-

fairscale/nn/pipe/__init__.py

@@ -22,5 +22,6 @@ from .async_pipe import AsyncPipe
 from .checkpoint import is_checkpointing, is_recomputing
 from .pipe import Pipe
 from .rpc import PipeRPCWrapper
+from .types import LazyModule
 
 __all__ = ["Pipe", "is_checkpointing", "is_recomputing", "LazyModule"]