Throw onerror instead of exception for AudioContext constructor

[mjwilson-google]

This affects #2435, but may not fix everything in that issue.

We discussed at TPAC 2024 that, because device enumeration is asynchronous, it will be easier for implementers and also more robust to use onerror to signal problems with the device at construction time.

This PR does the following:

• Move the sink validation to before system resources are acquired
• Change the sink validation algorithm to return boolean values
• Update both uses of the sink validation algorithm to use the new return value

---

Preview | Diff

[hoch]

I think this is what we discussed offline, but I am curious if there are places we might need to spot check,

[mjwilson-google]

Thank you for the quick review! I'm not sure if this is the best way to word this change, so any feedback or additional checking is welcome.

[padenot]

One small non-fonctional comment.

[mjwilson-google]

After the working group meeting this PR needs some rework, I will comment again when it is ready for review.

[mjwilson-google]

I think the current patch covers what we talked about at the meeting:

• Add private internal slot for sinkid, to hold the constructor parameter until we are sure the sinkid is usable
• Then we can update the javascript-visible slot with the valid ID once it has been opened
• If validation fails, fall back to default device, fire onerror, and do not transition to running
• For acquisition failure, always fire onerror, and we can remove the speaker-selection step since it will be handled in the validation step

I'm not sure about the naming of the private slot, and we also have two things called sending a control message to start processing (one in the constructor and one for setsinkid) but hopefully it's clear from context which one is which.