Address Faraday warnings on v0.x

[marcotc]

Fixes #965

When we added support for Faraday v1 in #906, a warning started to be emitted by Faraday versions < 1.0: WARNING: Unexpected middleware set after the adapter. This won't be supported from Faraday 1.0.
This did not affect the correct behaviour of instrumentation on Faraday < 1.0.

This PR addresses that warning by patching Faraday < 1.0 using our old approach, the one removed in #906, instead of the new patching used by v1.0.

The removal of the old approach was originally intended, as the new patching used for 1.0 also works for < 1.0, as stated, but this warning we are seeing is not desirable either.

I've added a new Appraisal category for contrib to allow us to test older version: contrib-old. I'm using it to test Faraday 0.17, while keeping our regular tests of the latest version of Faraday under contrib.

[delner]

The removal of the old approach was originally intended, as the new patching used for 1.0 also works for < 1.0, as stated, but this warning we are seeing is not desirable either.

I'm not so clear on this... changing the patching strategy in #906 to support 1.0 works for 1.0, but when present in versions < 1.0 causes that older version to raise a warning that it won't work in 1.0? Despite it actually working in 1.0?

[marcotc]

I'm not so clear on this... changing the patching strategy in #906 to support 1.0 works for 1.0, but when present in versions < 1.0 causes that older version to raise a warning that it won't work in 1.0? Despite it actually working in 1.0?

@delner this is accurate. I'm guessing they didn't predict exactly how 1.0 would be implemented, so the warning gets tripped even though it works.

[delner]

👍 Thanks for the fix, @marcotc !