Consolidate network download code

[dbent]

This consolidates much of our network download logic into Net, and standardizes behavior such that:

• We always try to use a WebClient first
• If that fails, we attempt to use cURL

Also, moves the cURL CA file resolution logic into Curl so that everything that uses it benefits from it.

[pjf]

Either I'm mis-reading the logs, or KerbalStuff is working fine with the default WebClient under Linux, which it didn't use to do and was one of the big motivators for having curlsharp support in the first place.

Trying to find more ways to test this. :)

[dbent]

I was slightly confused as well. What the code used to do was always try with CurlSharp, then catch a DllNotFoundException and then re-attempt with WebClient. So it is using CurlSharp on Linux, but silently failing on Windows (unless you have cURL installed) and being retried.

[dbent]

And to add, my changes standardizes the behavior so that it always attempts with WebClient first, then falls back to CurlSharp.

[pjf]

Looks like kerbalstuff supports a number of TLS 1.0 ciphers now, and I'm sure their absence was the reason we jumped to libcurl in the first place.

So I'm still testing, trying to find something that forces a failover to libcurl. :)

[pjf]

This seems to be working fine with me forcing errors. My only real concern is if mono's WebClient class ever results in a timeout due to bugs, which my head says shouldn't happen, but which my gut says has already happened.

However I can't find evidence to support my gut, so I'll be merging this shortly. :)