Add flake8 config

This enables a clean flake8 run, and is based on the existing flake8
configuration from Numba.

.flake8

@@ -0,0 +1,52 @@
+[flake8]
+ignore =
+    # Extra space in brackets
+    E20,
+    # Multiple spaces around ","
+    E231,E241,
+    # Comments
+    E26,
+    # Assigning lambda expression
+    E731,
+    # Ambiguous variable names
+    E741,
+    # line break before binary operator
+    W503,
+    # line break after binary operator
+    W504,
+max-line-length = 80
+
+exclude =
+    __pycache__
+    .git
+    *.pyc
+    *~
+    *.o
+    *.so
+    *.cpp
+    *.c
+    *.h
+
+per-file-ignores =
+    # Slightly long line in the standard version file
+    numba_cuda/_version.py: E501
+    # "Unused" imports / potentially undefined names in init files
+    numba_cuda/numba/cuda/__init__.py:F401,F403,F405
+    numba_cuda/numba/cuda/simulator/__init__.py:F401,F403
+    numba_cuda/numba/cuda/simulator/cudadrv/__init__.py:F401
+    # Ignore star imports, unused imports, and "may be defined by star imports"
+    # errors in device_init because its purpose is to bring together a lot of
+    # the public API to be star-imported in numba.cuda.__init__
+    numba_cuda/numba/cuda/device_init.py:F401,F403,F405
+    # libdevice.py is an autogenerated file containing stubs for all the device
+    # functions. Some of the lines in docstrings are a little over-long, as they
+    # contain the URLs of the reference pages in the online libdevice
+    # documentation.
+    numba_cuda/numba/cuda/libdevice.py:E501
+    # Ignore too-long lines in the doc examples, prioritising readability
+    # in the docs over line length in the example source (especially given that
+    # the test code is already indented by 8 spaces)
+    numba_cuda/numba/cuda/tests/doc_examples/test_random.py:E501
+    numba_cuda/numba/cuda/tests/doc_examples/test_cg.py:E501
+    numba_cuda/numba/cuda/tests/doc_examples/test_matmul.py:E501
+    numba_cuda/numba/tests/doc_examples/test_interval_example.py:E501