Allow test_driver.bless to work when running manually

[foolip]

Tests using test_driver.bless now currently always fail when running
them manually, because the default test_driver_internal.click
rejects with "unimplemented".

Mute the rejection if the session isn't controlled by WebDriver, which
we check with navigator.webdriver.

[foolip]

This is an alternative to #11173, and only fixes test_driver.bless.

[jgraham]

I'm somewhat concerned about the use of the webdriver property here; although it's possibly fine it's unclear how future-proof it is to assume that all automation is webdriver-driven.

[foolip]

I'd be happy to use any other signal we have that we're being controlled by wptrunner. Do we have something else we could use?

[jugglinmike]

That's the question which is blocking gh-11173.

[foolip]

The answer is straightforward I think: use navigator.webdriver, add a test to /infrastructure/assumptions/ for it, and refine if there's a problem. The consequence of getting it wrong is just to change the failure mode from fast fail to timeout.

[gsnedders]

@foolip does this work for run-webkit-tests and Marionette and CDP and other non-WebDriver runners?

[foolip]

@gsnedders, no, but there we'll replace test_driver_internal.click anyway. (Moving this sort of thing into test_driver_internal.* might make sense.)

[foolip]

@jugglinmike, do you intend to get #11173 landed? If that's not imminent I'd like to land something simple like this PR that makes manual testing possible.

[jugglinmike]

I'd like to land gh-11173. I've just updated it in an attempt to satisfy the remaining concerns.

[foolip]

All right, reviewing that.