Introduce start session algorithm

[beaufortfrancois]

This PR addresses concerns raised in #135 (comment) and #126 (comment):

• It makes sure start() and start(MediaStreamTrack audioTrack) throw if document is not fully active:
• Request permission to use microphone is called only for start()
• A new "start session algorithm" is introduced to formalize how things work when speech recognition starts.

The following tasks have been completed:

• Updated web-platform-tests: https://chromium-review.googlesource.com/c/chromium/src/+/6237075

Implementation commitment:

• Blink: (link to issue)
• Gecko: (link to issue)
• WebKit: (link to issue)

---

Preview | Diff

[beaufortfrancois]

@evanbliu @padenot @youennf Let me know if that looks good to you

[beaufortfrancois]

Note that I used UnknownError to match WebKit that already shipped Web Speech.

FYI Firefox uses InvalidStateError instead but did not ship Web Speech. So even though, I'd personally be in favor of InvalidStateError, I picked UnknownError so that web developers already catching this properly based on Safari don't have to update their code.

[youennf]

I think webkit could align to whatever deemed appropriate, this is an edge case so this is likely not a web compat issue.
I also prefer InvalidStateError.

[beaufortfrancois]

Since you and @evanbliu also prefer InvalidStateError, I've updated this PR to use it.

[beaufortfrancois]

Quick question for you folks!

When adding web platform tests for this in https://chromium-review.googlesource.com/c/chromium/src/+/6237075, I used frameWindow.SpeechRecognition = frameWindow.SpeechRecognition || frameWindow.webkitSpeechRecognition; so that it works in Safari and Chrome BUT technically, it's not a spec-compliant as webkitSpeechRecognition is not a thing. I still think there's value testing implementation with webkitSpeechRecognition but I'd love your thoughts on this.

[evanbliu]

Blink owners have requested that we drop the "webkit" prefix from the Web Speech API implementation in Chrome. We'll have to maintain backwards compatibility indefinitely, but eventually we'll probably pretend like it doesn't exist.

[beaufortfrancois]

I've addressed @evanbliu's feedback and resolved conflicts. Let me know if there's more to address or if we can merge it.