fix concurrent recording by using a waitgroup

[stefanotorresi]

Fixes #34

I should have done this since the beginning: to simplify, RecordConcurrently only sent the first error that occurred among all the concurrent metric recording functions, returning as early as that error occurred.
This could cause some of the goroutines to leak outside the life-cycle of Collect, causing a panic when trying to send on a closed metrics channel.

To be fair I didn't know that, once Collect is done, the metric channel would be closed.

The RecordConcurrently implementation now uses a Waitgroup to wait for all the recording functions, avoiding any of them to leak, and returns all the errors in an array, instead of just the first.

[stefanotorresi]

I had to add the fmt and vet checks to the test target in the Makefile because I realise that I always forget to run them, and I always use make test. Actually let me do this elsewhere.

[MalloZup]

lgtm