There is a single implementation of it, so inline it.

Also make sure we only initialize device change listeners once.

We require navigator.mediaDevices.enumerateDevices to exist.

It appears that when the gDM flow is used we do not set properly the sourceId of the shared stream.  When the user has multiple screens we use the sourceId to identify the display that is currently shared. Therefore in electron we can't identify the display that is used and we fail.

Technically anyone can use it as long as they provide connection
options, but we don't need any of that for JaaS.

It's a short-hand for leaving the conference and then disconnecting the
underlying connection.

In the general case (no breakout rooms) this is likely a preferred way
to handle leaving, since there is no good reason to keep the connection
open afterwards.

The bridge is now able to use DD headers for VP8 and VP9 as well so there is no need to remove it from remote desc when the call switches from AV1 to VP8/VP9.

Disables AV1 for FF since AV1 is not properly supported (it doesn't support SVC and it doesn't send temporal layers with simulcast).

Added a testing config flag so that we can easily enable it and test it.

Disables VP9 for FF since the simulcast  is disabled with FF flag by default. Introduces a few fixes so that if the FF flag is enabled simulcast works. This should also help us to be ready to enable VP9 once the  FF config flag for simulcast for VP9 is enabled by default.

For FF 136+ where DD extension is supported,  removing DD extension from the SDP will lead to broken scalabilty (the temporal scalability information will be missing and also DD extension header wont be included).

Endpoint based constyraints support was removed 2 years ago. This check
can go now.

* move rtcstats init

* tame lint

After enabling E2EE for the first time, keep the media transform
inplace, but make it a no-op.

This simplifies the flow a little since it doesn't require a media
session restart.

Way back when, there were performance concerns when having a transform,
which is what prompted us to implement the restar mechanism, but a lot
has changed since and both audio and video frames don't need to jump
threads these days. We are still not enabling an empty transform at all
times out of precaution.

* feat(startMutedPolicy): Uses metadata for startMuted policy.

* squash: Make sure we add boolean.

* squash: Fix lint.

* squash: Ignore policy when set locally.

* squash: Adds a comment.

* squash: Fix lint errors.

Initializing JitsiMediaDevices doesn't rely on any settings / config
passed on to JitsiMeetJS.init, so make sure it's initialized only once.

This is only permitted when testing.testMode=true in config.js

This way we can use it both in Webpack and TS builds.

This makes it use the ESM build rather than the UMD one.

- It can throw exceptions
- It doesn't return false on things which are not Number, just on things
  that are not NaN

Fixes: https://github.com/jitsi/lib-jitsi-meet/issues/2743

The end goal is to fix our types to make typedoc generated docs as
useful as possible.

The generated docs will be published to GH pages on every push.

In addition:

- Remove old doc pointer
- Replace example with a more comprehensive one

When joining if there was a modification of presence just before sending the presence in the same millisecond we send later a second presence. While testing we see modifications from e2ee which cause the duplicate presence.