Improve auto-detection to avoid long web request timeouts

[mjcheetham]

As mentioned in #364, during the host provider auto-detection when GCM makes a HTTP call to the remote, this can sometimes hang until a long timeout period has elapsed (1 minute?).

We should improve the "2nd level" auto-detection in a few ways:

1. Add fast-path identification of other host providers from the remote URL hostname
2. Add more heuristics for common on-prem remote URLs like we do for github.<domain>.<tld> => GHES/AE
3. Set a short timeout for the HTTP request
4. Parallelise the HTTP request with other "1st level" detection
5. Cache the results of the identification request(?)
   a. Prompt the user to remember (set credential.provider) the detected provider(?)

[Varriount]

This is worse than just "long requests" - if one is pushing to a site that doesn't support Microsoft Authentication, and the credential provider isn't set to "generic", the credential manager will eventually be killed by Git, causing Git to fall back to asking for credential input from the user. This essentially makes Git Credential Manager's default configuration broken for any site that isn't GitHub.

[fourofspades]

This problem is killing us, 2+ seconds for every git request. Was this even tested? The documentation is also poor. How do I specify a server without a hostname, i.e. by ip address, this is unclear.