fix(navigation): do not count random failures as navigation cancel

[dgozman]

In WebKit, we sometimes get the following:

• page starts a navigation in some frame;
• we issue page.goto();
• new navigation commits;
• old page is still alive and cancels the frame navigation;
• while waiting for 'Page.frameNavigated' on the frontend from the new target, we receive 'Browser.provisionalLoadFailed' for the frame navigation from the old target;
• this counts as cancel for the new navigation, although it was clearly a cancellation of some other navigation.

Wasn't able to write a reliable test. Any ideas are welcome! Fixes #1044.

[yury-s]

As for testing this, what if we have two consecutive .goto and check that first is cancelled by the second while the second succeedes? The first request could be deferred on the server for better reliability

[JoelEinbinder]

code looks good. Idk about the test.

[dgozman]

As for testing this, what if we have two consecutive .goto and check that first is cancelled by the second while the second succeedes? The first request could be deferred on the server for better reliability

We have 'should fail when canceled by another navigation' which does exactly that. Unfortunately, the issue is slightly different here: first navigation is canceled and that mistakenly "cancels" the second one.