Fix various leaks

[Arno500]

When running fishymetrics in our own infrastructure we noticed that the memory usage was greatly increasing (at a steady pace)
as you can see here:

Left: deployed as-is, middle: after SystemD memory accounting and limiting, right: with the fix

The specificity of our configuration is that we are scrapping around 6000 servers on this instance (there are actually two instances load balanced, so let's say 3000 averaged) and that a non negligible portion is in error(either because of a broken BMC firmware than cannot properly be parsed by Fishymetrics, or just because they are unreachable). There was two issues:

1. A channel is awaited during the scrapping, but if, for whatever reason, there was a fetching/parsing error during the handling of the task, then we have a leaking go routine, repeating at every scrap
2. Some body were not closed when requesting (due to the structure of the code) and in some cases the body was not consumed, preventing the connections from being reused (see the associated commit, Prometheus had a similar issue in the past)

With those two fixes, we get the memory usage under 3GB for those 3000 servers, which seems much more reasonable to me :)

[CLAassistant]

All committers have signed the CLA.

[Arno500]

(I'm currently trying to sort up my pro account to be able to sign the CLA)

[ibrahimkk-moideen]

@Arno500 pls look into the changes requested.

[ibrahimkk-moideen]

LGTM.