[CircleCi]: use more fallbacks in case of a failure

[wilzbach]

• adds the fallback to nightlies in case of a failure to download the install script (see also the regarding Travis PR)
• fix the soft fallback to master in case of an overuse of the GitHub API has e.g. occurred at Rework MersenneTwisterEngine using Ilya Yaroshenko's faster algorithm #5011 (with the -e option the bash script immediately exits on an error, hence the || echo trick)

[DmitryOlshansky]

I know nothing about CircleCI automation but this looks sensible

[wilzbach]

I know nothing about CircleCI automation but this looks sensible

It's behaves like ever other CI provider: on every synchronize event of a PR, they trigger a new build.
It's used here to do static code analysis via Dscanner (the style Makefile target)
Moreover, it runs all unittests with coverage logging enabled and uploads the results to CodeCov.

All in all the build is executed with the circleci.sh Bash script which then does a bit of bootstrapping. Most importantly this is

1. cloning the DMD and Druntime repo
2. merging the upstream Phobos branch (this requires an API call to the GitHub API which can run into rate limits)
3. download and install a DMD compiler

As mentioned above this adds fallbacks for download errors occurring at step 2 or 3 of the bootstrap phase.

[DmitryOlshansky]

LGTM

[DmitryOlshansky]

Auto-merge toggled on