Add node_filesystem_device_errors_total metric

[discordianfish]

This metric is the total number of errors occurred when getting stats
for the given device.

This closes #325

[SuperQ]

It'd be nice if there was a test for this, not sure how easy that would be.

[discordianfish]

@SuperQ Yes more tests would be nice in general, but not sure how to do in such case. Mocking up those things seems silly, but to be honest I'm not a testing expert. Maybe we could actually run the node-exporter on some frozen VM state but that seems a big project. Either way, I'd say it's not a blocker to merge this.

[SuperQ]

Testing would involve mocking syscall.Statfs() to return and error, and calling GetStats(). Then checking to see if the counter increased. Not terribly hard, but not super easier either. I think this is fine for now.

[discordianfish]

@SuperQ Yes but that would mean turning this call into some interface to mock it. Not sure it's worth it, given how this would apply to so many things in the exporter. That why I was thinking about using a frozen VM or something. But first things first.