Keep AudioContext object and add WILL_START dictate state

[compulim]

Fixes #2516.
Fixes #2518.
Fixes #2519.

Changelog Entry

• Fixes #2516. Disable microphone input for expecting input hint on Safari, by @compulim in PR #2517 and PR #2520
• Fixes #2518. Synthesis of bot activities with input hint expecting, should be interruptible, by @compulim in PR #2520
• Fixes #2519. On Safari, microphone should turn on after synthesis of bot activities with input hint expecting, by @compulim in PR #2520

Description

Final mile for "input hint expecting" feature.

This will fix 3 issues, regarding interruptibility and compatibility with Safari.

Specific Changes

• Introduced a new dictate state called WILL_START
  • After synthesizing all pending activities, this dictate state will turn into STARTING, which will start the speech engine
  • This improve interruptibility of activities
• Keep the same AudioContext object after first creation
  • AudioContext will receive blessing after the end-user consent on the privacy dialog
  • The blessed object can be used for recognition without additional user interactions (click or tap)
  • The blessed object should be kept across multiple recognition sessions

---

• Testing Added
  • No tests are currently added, we will add manual steps to the release checklist

[coveralls]

Coverage decreased (-0.1%) to 63.345% when pulling b9adae4 on compulim:fix-last-speech-bug into 69b4a13 on microsoft:master.

[corinagum]

Tests after build