The indexes returned in recording events from BBB refer to positions
within a UTF-16 encoded string. Rather than attempt to untangle this in
the server (which might have a performance cost), it's easier to switch
the caption processing code to operate in UTF-16 encoding as well to
make it work consistently.

The PyICU library provides a UnicodeString type which is a UTF-16 string
similar to Java and JavaScript, but which supports all the python
indexing methods. It's fairly straightforwards to swap it in in place of
the types used previously, and works natively as an input to the ICU
line break iterator too.

Fixes #10531

The previous implementation of the BigBlueButton.execute method runs the
process with separate stdout and stderr streams. It first reads all of
the output from stdout, then reads all of the output from stderr.

This can cause a deadlock if the process writes a lot of data to stderr.
The IO buffer for stderr could fill, blocking progress. But since it
hasn't closed stdout, the ruby script is still waiting on a read to
stdout.

Switch to an execution method (using IO.popen) that allows combining
stdout and stderr into a single stream, eliminating the issue.

On my server 2.3 alpha, the method metadata_for(meeting_id) gives back {}
(empty Hash). Thus "return if meeting_metadata.nil?" does not occur.

Does @redis.hgetall give {} instead of nil, even though there is a comment in
node_modules/redis/lib/utils.js "hgetall converts its replies to an Object. If
the reply is empty, null is returned"???

This commit fixes an issue with reading and writing files.
File.open is used which means that a file will remain open
unless explicilty closed or the program exit. This doesn't work
for an NFS mount as the scripts try to "rm -rf" when the file
is still open. This commit fixes that by replacing all .opens
with .reads

There's nothing reasonable we can do here if the font size is missing.

After the last segment of a recording has been archived, it is safe to
delete the original files that were recorded by the various media
handling components. This patch deletes the freeswitch audio files and
kurento webcam/screenshare files after they have been archived.

It needs to be used in combination with some changes to users/groups and
directory permissions to allow the `bigbluebutton` user to delete files
that it could previously only read.

Fixes #8252

Bumps [loofah](https://github.com/flavorjones/loofah) from 2.2.3 to 2.3.1.
- [Release notes](https://github.com/flavorjones/loofah/releases)
- [Changelog](https://github.com/flavorjones/loofah/blob/master/CHANGELOG.md)
- [Commits](https://github.com/flavorjones/loofah/compare/v2.2.3...v2.3.1

)

Signed-off-by: dependabot[bot] <support@github.com>

Add an example to enable generating an mp4 file (Apple device compatibility)
Switch the webm generation to use the faster single-pass encode by default,
since 2-pass is much slower and more cpu intensive.

The wav files from the freeswitch recordings are huge, uncompressed.
There's no reason to keep them as wav, we can compress them in the
archive step.

Someone on the mailing list had some recordings which were using the 2.0
playback, but were missing the deskshare.xml file (which should always
be present for 2.0… strange). It's safe to continue loading the recording
playback if the deskshare.xml file is not found, the recording will just act
as if there were no deskshare start/stop events.

Previously the setMediaSync function was only called after the deskshare
loaded, but by moving it to run after all media loaded, it now runs even
on recordings that didn't have deskshare. Make it do nothing (return early)
in that case.

If the secondary media loaded before the main media, it would run the
"setMediaSync" function before the main media player was setup. As a
side-effect of setting up the main media player, all of the event handlers
added by the setMediaSync function are detached, and so the secondary
media never starts playing.

Move the call to setMediaSync to after the media-ready events for all
media have fired, so that it can reliably attach the event handlers.

Bumps [rubyzip](https://github.com/rubyzip/rubyzip) from 1.2.3 to 1.3.0.
- [Release notes](https://github.com/rubyzip/rubyzip/releases)
- [Changelog](https://github.com/rubyzip/rubyzip/blob/master/Changelog.md)
- [Commits](https://github.com/rubyzip/rubyzip/compare/v1.2.3...v1.3.0

)

Signed-off-by: dependabot[bot] <support@github.com>

Adapt the screenshare processing script to changes to the api for events handling in
the recording-and-playback core libs.

This reverts a bbb-specific customization made in the jquery.acornmediaplayer.js
file: it's restored to what the upstream player did. I can't find any explanation
for why this change was made in the first place? Reverting it doesn't seem to
cause any playback issues (Popcorn still works, in particular).

Bumps [nokogiri](https://github.com/sparklemotion/nokogiri) from 1.10.3 to 1.10.4.
- [Release notes](https://github.com/sparklemotion/nokogiri/releases)
- [Changelog](https://github.com/sparklemotion/nokogiri/blob/master/CHANGELOG.md)
- [Commits](https://github.com/sparklemotion/nokogiri/compare/v1.10.3...v1.10.4

)

Signed-off-by: dependabot[bot] <support@github.com>

Found another case where the html5 client was passing through control
characters, in the original presentation name field.

Rather than play whack-a-mole with different fields which may eventually
get poorly sanitized user data, apply the control character filtering
to all properties.

Adjust the character range to do the following:
* Allow horizontal tab (0x09), it's not problematic.
* Disallow control characters in the range 0x1A-0x1F. Probably missed by accident.