It seems to work well with current ecosystem and stackage
has just introduced 8.6.3 as part of lts-13.1:
  https://www.stackage.org/lts-13.1

In most cases, when we run a subshell with our own function
and assign the result to a variable, we also need to check that
the variable is non-empty, since 'die' and 'edo' don't propagate
to the parent shell.

In some cases, non-emptiness is handled in other ways or is not
fatal.

If a GHC is already installed, we shouldn't treat it as an
error.

Note that the packages `ncurses-term` and `gmp-devel` are required to run ghc.

Amazon Linux introduces itself as

```
andrea:~ $ cat /etc/os-release 
NAME="Amazon Linux AMI"
VERSION="2018.03"
ID="amzn"
ID_LIKE="rhel fedora"
VERSION_ID="2018.03"
PRETTY_NAME="Amazon Linux AMI 2018.03"
ANSI_COLOR="0;33"
CPE_NAME="cpe:/o:amazon:linux:2018.03:ga"
HOME_URL="http://aws.amazon.com/amazon-linux-ami/"
```

Partly "fixes" #7

"fast_finish: true" is only effective when you have jobs in your
build matrix that you allow to fail, which isn't the case here.

https://blog.travis-ci.com/2013-11-27-fast-finishing-builds

Fixes #50

Resolves https://github.com/haskell/ghcup/issues/33