Add -Wimplicit-fallthrough

[Quuxplusone]

Clang doesn't enable this warning by default, so we pass it explicitly. GCC pre-GCC 7 doesn't recognize the flag, so we can't pass it for GCC. GCC 7 enables it by default as part of -Wextra.

Fixes #68.

[Quuxplusone]

Successfully rebased on master. The remaining Github CI test failures are due to brew update timing out (what should we do about this recurring problem, @still-flow?) and the MinGW static initialization order fiasco (which I keep meaning to investigate — I mean the whole idea of initializing a global to geometry[ginf] seems wrong, because that value will be out-of-date immediately — but I haven't really tackled it yet).

So @zenorogue ping!

[still-flow]

brew update timing out (what should we do about this recurring problem, @still-flow?)

Not much, I'm afraid. Looks like an intermittent issue to me: there are successful MacOS builds along with a couple failed ones in the same run, and I don't recall anything too different in how homebrew is handled from job to job -- after all, each job is supposed to run in a separate VM starting from the same image.

Some superficial searching yielded "solutions" mentioning disabling proxy in git config or overly-keen firewalls/antiviruses, but the responses were inconclusive. But given what I've said above, it's probably worth to report this to Github support.

UPDATE: actually, I'm not quite sure that brew update is all that necessary. Don't know the specifics, but it might be the case that the package database is already present in the VM image, even if slightly outdated. So removing this step might help?