[azure-core] Small fixes for aiohttp

[annatisch]

There are 3 fixes in this PR:

• Aiohttp does not define an error for request read timeouts - instead it relies on the async future to timeout. The patch I have here catches the asyncio.TimeoutError and raises the ServiceResponseError - however I'm not entirely sure this correct - given that there may be other reasons the future times-out. Either way, whichever is raised here, the retry policy must be updated to work with it.
• Aiohttp ClientResponse object has a "status" attribute rather than the "status_code" used by Requests.
• Aiohttp uses a single string for the content-type headers rather than the list used by Requests.

[adxsdk6]

Can one of the admins verify this patch?

[annatisch]

Added another fix to this PR - found by @iscai-msft
There was an issue with a couple of the async transports inheriting the synchronous sleep function which then failed when called as a coroutine (e.g. in the async retry policy).

Sorry to add another feature @bryevdv ....

[annatisch]

@chlowell - I remember at one point we put handling into azure-core because aiohttp (iirc) was returning the content-type header as a list rather than a string. This no longer seems to be the case... can you recall the particulars? Or am I getting completely confused with something else?

[lmazuel]

So CI, is doing something really weird:

• All Python 3.5 tests are passing for "identity" CI
• Python 3.5 Windows passes for all tests, but linux fails for "identity" tests
• Everything else passes in all platform and Python

I believe it's an issue in the CI :(. Re-running just the job (sigh)