This is an averaged, normalized version of a number of examples of each digit in the odometer.
It does not include the last digit which looks different.
These prototypes are used to recognize the digits in buscam data.
due to not including the map data in the hash calculation.
This is only relevant for streams with map data, which does not include twitch or youtube URLs.
since they can't do any other quality, but we still want to be able to set other qualities
for twitch streams.
Really qualities should be per-channel but I'm being lazy.
Adds very simple youtube stream support where we only ever use the "best" quality,
which we call "source" for consistency with twitch.
We use yt-dlp to do the heavy lifting of getting the playlist url out of youtube.
In some formats, most notably DASH, there is a "initialization data" that is required
in order to play the segment. The data is common to all segments so it is served as a seperate URL
under EXT-X-MAP. However, redundant copies of this data are benign and it's very small, so
we just put it in front of EVERY segment so that we can play every one independently (but
concatenating them still works).
We use a very simple cache to avoid downloading it again for every segment.
This is mainly a problem with groups_by_shift which will be set
even for negative hours that aren't on the schedule.
This causes notification spam as people move in and out of groups.
This leads to delayed JOIN/PARTs not being put in their proper batch
as it's already been closed. In fact, since each message is re-opening a batch from
more than MAX_SERVER_LAG seconds ago, each message becomes one batch.
* Join every channel on connect
* Move the "wait for initial ROOMSTATE" logic into the main loop and make it per-channel
* Make batch keys (channel, time) instead of just time
For now the CLI doesn't actually allow you to run chat_archiver in this mode,
it always calls the implementation with a 1-element list of channels.