In order to support XMPP authentication, we'll need the message
accompanying the error and carried by lib-jitsi-meet's
CONFERENCE_FAILED in the redux store. We already carry the message in
the redux action and we've got the error in the redux store.

In order to accommodate the requirements of the work on supporting XMPP
authentication on mobile/react-native, make dealing with Dialog a
little more generic and a little easier.

Filmstrip only mode displays a device selection dialog that
does not have access to the redux/connect. However, the current
VideoTrack extends from AbstractVideoTrack, which assumes a
redux connection. The fix is to move video display logic into
a separate component and have device selection use that, while
the existing VideoTrack remains connected to redux but
uses the new video display component.

Instead of going through the Avatar object, call the getAvatarURL
directly so that the code flows consistently use the participant
representation within redux.

In certain scenarios we do need to import files instead of features but
even then it is our coding style to not write the file extension.

We broke external_api.min.js by importing react/features/util which
imported react/features/base/lib-jitsi-meet.

1. To reduce the risks of such a breakage until we add
   external_api.min.js to the torture tests, import as little as
   possible in modules/API/external/external_api.js.
2. Use the global JitsiMeetJS on Web in react/features/base/util.

The error is stored in the redux store in base/config so other components can
consult it. It is also broadcasted as a new event in the external API for the
SDK.

- Remove references to the model ContactList.
- Replace ContactListView with an empty element for attaching
  the React Component ContactListPanel, which has the same
  features as the old ContactListView.
- Create new selector for getting non-fake participants for
  ContactListPanel's props.
- Create a ParticipantCounter component to place in the contact
  list button. Previously ContactListView updated that but now
  it's a react component hooked into the participant state.
- Remove pub/sub that was used only by ContactListView.

Implement a React Component which displays children as an overlay of
local video. The WelcomePage implemented such a component inside of it
among other WelcomePage-specific logic so I split
LocalVideoTrackUnderlay out of it. The new Component is used on the
BlankPage which may be displayed in the future not only when the
WelcomePage is disabled but also when there are long running network
requests, for example.

This only helps iff there is a short transient network error which prevents the
configuration from being loaded. In such case, use the cached version in
localStorage, which may not match the shard, but it's (probably!) better than
nothing.

In case there is no Internet connectivity, an error will be produced as soon as
the XMPP connection is attempted anyway.

It's built on top of React Native's AsyncStorage. They have differing APIs, so
we implement a synchronous API on top of an asynchronous one. This is done by
being optimistic and hoping that operations will happen asynchronously. If one
such operation fails, the error is ignored and life goes on, since operations
are performed in the in-memory cache first.

Note to reviewers: LocalStorage.js lacks Flow annotations because indexable
class declarations are not yet supported:
https://github.com/facebook/flow/issues/1323 and yours truly couldn't find a way
to make the required syntax work without making it unnecessarily complex.

It'snot the end of the world, we can move on.

This patch loads the config later than we used to, that is, only once we
know the room the user is about to join.

Due to architectural limitations in lib-jitsi-meet, it needs to be
initialized with a configuration in order to properly function. This is
unfortunate because we need to create a video track in the welcome page,
but don't know the room (hence no config) yet. In order to circumvent
this problem an empty configuration is used, which is later swapped with
the appropriate one, once loaded.

Some interesting side-effects of this change are a perceived speed
increase when the app starts or a conference is hangup. They are both
due to the fact that no config needs to be fetched from a remote server
in those cases.

For example, while config.js and other files are being loaded before the
navigation to Conference is feasible.

In order to load the configuration from the shard that will actually
host the conference, it's imperative that we add the room= query
parameter:

https://meet.jit.si/config.js?room=example

This implies a departure from our current model, where the config is
discarded if the domain for the next conference is different, but kept
otherwise.

* feat(filmstrip): show thumbnails with toolbar and on hover

* squash: reduce verbosity of logic for when to display

* squash: remove check for fake participant

Before fake participant (youtube video) would make the filmstrip
always displayed. However, youtube videos already dock the
toolbar, so filmstrip will remain displayed, so the check is
redundant.

* squash: change mouse hover listener targets

* feat(analytics): move to React

The analytics handlers have been moved to JitsiMeetGlobalNS, so now they are
stored in `window.JitsiMeetJS.app.analyticsHandlers`.

The analytics handlers are re-downloaded and re-initialized on every
lib-jitsi-meet initialization, which happens every time the config is changed
(moving between deployments in the mobile app).

* Adds legacy support for old analytics location.

We have already made the implicit decision not to pursue what the
comment describes. If we ever revisit it, it probably won't be handled
where the comment is anyway.

The torture tests were looking for the anchor tag within each
button. However, that anchor could get blown away from a react
re-render. So instead, expose a way for the torture tests
to find the root node of the button.

I'm not saying that the two commits in question were wrong or worse than
what I'm offering. Anyway, I think what I'm offering brings:

* Compliance with expectations i.e. the middleware doesn't compute the
next state from the current state, the reducer does;

* Clarity and/or simplicity i.e. there's no global variable (reqIndex),
there's no need for the term "index" (a.k.a "reqIndex") in the redux
store.

* By renaming net-interceptor to network-activity feels like it's
preparing the feature to implement a NetworkActivityIndicator React
Component which will take on more of the knowledge about the specifics
of what is the network activity redux state exactly, is it maintained by
interception or some other mechanism, and abstracts it in the feature
itself allowing outsiders to merely render a React Component.

It will only be shown when there are ongoing network requests.

Works only for XHR requests, which is the only network request mobile performs
(WebRTC traffic aside). The fetch API is implemented on top of XHR, so that's
covered too.

Requests are kept in the redux store until they complete, at which point they
are removed.

The global.performance.now polyfill is removed since it's polyfilled by
react-native-callstats.

* ref(1-on-1): move remote visibility to a selector

Derive whether or not remote videos should display using a selector
to look across different states. A selector was chosen over using
memoized selectors (reselect) or subscribers as a first step
approach, avoiding additional mutations caused by a subscriber
updating the filmstrip state and avoiding additional api overhead
introduced by reselect.

* rename selector