fix($httpBackend): use ActiveX XHR when making PATCH requests on IE8

[IgorMinar]

IE8's native XHR doesn't support PATCH requests, but the ActiveX one does.

Closes #2518
Closes #5043

[caitp]

Some quick ways to get tests passing again without getting rid of the noxhr minerr:

• https://gist.github.com/caitp/ff79ca952810ad0e867f
• https://gist.github.com/wesleycho/3e204bbee09b5a9103e4 << Has the advantage of being much more concise, but I don't know if we necessarily want to throw away the different versions of XMLHTTP if they're available. An IE guru would know more about that.

[mgol]

@caitp jQuery 1.x uses only the simple form:
https://github.com/jquery/jquery/blob/3140d3bdd3a520e98c5178b04eba68ce16dacf46/src/ajax/xhr.js#L179-183
and it works fine all the way back to IE6 so I don't think such granularity is needed.

[IgorMinar]

silly me! I didn't realize that this PR didn't pass our tests.

sorry @caitp for wasting your time.

the noxhr error is thrown because I removed the only instance of the error from the code, but not the doc for this error and the documentation generation code doesn't expect to come across an error doc for an error that is not being used.

as @mzgol it should be fine to use the simplified version of the xhr instantiation. I checked with jquery source as well as some other places and this short form is legit.

[IgorMinar]

ok, I pushed a new version now. it should be good to go.

[IgorMinar]

travis/saucelabs/karma is not having a good day today, but our ci server says that this is good to go: http://ci.angularjs.org/job/angular.js-igor/1159/

[caitp]

I don't consider it a waste of time at all --- this appears, as you said, to be a widely adopted solution, but I would still prefer to see tests where the browser can make real requests, just so that we can verify that the actual transport stuff is working as expected across all supported browsers, and have some kind of alert when if/it breaks.

I don't know if there's a good infrastructure for that right now, but maybe it could be tacked on as an afterthought later, I'm not sure.

But if this is good to merge, then awesome

[IgorMinar]

we don't have a good infrastructure for this kind of test right now. we could write a e2e test and add it to our build system, but it's not worth doing for this particular issue since it's an IE8 thing and you know how we feel about IE8.

instead I manually tested this and verified that it works as expected and in fact I found an unrelated issue that I also fixed in fd9a03e. IE8 is such pleasure to work with...