Do not pip install xgboost if already installed

[fritzo]

Addresses #100

This adds a workaround to setup.py to avoid pip installing xgboost if it can already be imported. This works for me when creating a new conda env, e.g.

conda create -n my_env python=3.6
conda activate my_env
conda install -c conda-forge xgboost
cd causalml
pip install -e .  # works after this PR

I am unsure whether this messes up the requirements pypi requirements, i.e. when releasing. This may only work when installing locally via pip install ..

[fritzo]

Closing because I believe this messes up requirements when pip uploading. I'll reopen if we can find a solution. cc @neerajprad