Allow pip install without NumPy or Cython

[jjhelmus]

Allow fastparquet to be installed when NumPy and Cython are not already
installed. pip attempts to run python setup.py egg_info to detemine the
install requirements. Allow this command and other non-build related commands
to be run without importing numpy or cython.

closes #170

[martindurant]

So what happens when you pip-install, but numpy or cython is not around?

[jjhelmus]

With this change, pip will try to download and install the requirements in the setup.py file including NumPy and Cython if they are not already installed. This will typically trigger builds from source which can take a long time and may fail, but it is the expected behavior.

For example:

$ conda create -y -n fp_test python=3.6 pip setuptools
Fetching package metadata .........
Solving package specifications: .

Package plan for installation in environment /home/jhelmus/anaconda3/envs/fp_test:

The following NEW packages will be INSTALLED:

    openssl:    1.0.2l-0      defaults
    pip:        9.0.1-py36_1  defaults
    python:     3.6.1-2       defaults
    readline:   6.2-2         defaults
    setuptools: 27.2.0-py36_0 defaults
    sqlite:     3.13.0-0      defaults
    tk:         8.5.18-0      defaults
    wheel:      0.29.0-py36_0 defaults
    xz:         5.2.2-1       defaults
    zlib:       1.2.8-3       defaults

#
# To activate this environment, use:
# > source activate fp_test
#
# To deactivate this environment, use:
# > source deactivate fp_test
#

$ source activate fp_test
(fp_test) $ pip install .
Processing /home/jhelmus/workspace/fastparquet
Collecting pandas (from fastparquet==0.1.0)
  Using cached pandas-0.20.2-cp36-cp36m-manylinux1_x86_64.whl
Collecting numba>=0.28 (from fastparquet==0.1.0)
  Using cached numba-0.33.0.tar.gz
...
Installing collected packages: six, python-dateutil, pytz, numpy, pandas, llvmlite, numba, ply, thriftpy, py, pytest, cython, fastparquet
  Running setup.py install for numba ... done
  Running setup.py install for fastparquet ... done
Successfully installed cython-0.25.2 fastparquet-0.1.0 llvmlite-0.18.0 numba-0.33.0 numpy-1.13.0 pandas-0.20.2 ply-3.10 py-1.4.34 pytest-3.1.2 python-dateutil-2.6.0 pytz-2017.2 six-1.10.0 thriftpy-0.3.9

[martindurant]

OK, this seems to be a reasonable solution - if perhaps I don't expect it to be used very often. I hope not to see follow-on issues from compilation error of numpy itself!