Dynamic source tracks.

[toots]

There still are reports of issues with source.tracks forcing sources to compute their content_type before we had a chance to assign their frame_type.

For reference:

• frame_type is a type representing the source's content as computed during type unification, e.g. {audio=pcm(stereo)}, {video=canvas, audio=pcm(5.1)} etc. This value can be incomplete if no full restrictions are put on the source content.

For instance, in:

s = blank()
output.dummy(s)

s can be any content because output.dummy does not care about the source's content. So, its frame_type is 'a

• content_type is a fully realized content value. It is calculated from the frame_type, adding default values.

For instance, a source with PCM audio without a specified number of channels, {audio=pcm('a)} is assigned a content type of {audio=pcm(stereo)}.

The problem is that there can be a competition between requesting the source's content_type and assigning the frame_type. If content type is requested before all the frame_type have been assigned then the result can be incompatible, e.g. assigning a stereo channel when the source should have had 5.1 channels.

The most typical case is when computing source.tracks. In the current implementation, the tracks are all extracted immediately at evaluation time which forces the computation of the source's content type right away.

In this PR, we introduce a delayed Source_tracks runtime value. This value can be passed in-lieu of the actual record of source tracks and evaluated only when needed, e.g. in the muxer operator.

In order to make things work, we need to also introduce a notion of dynamic_methods so that we can do the following:

# s = blank();;
s :
  source('A)
  .{
...
  } where 'A is a set of internal tracks =
  <source(id=blank, frame_type=something that is a set of internal tracks>.{...}
# t = source.tracks(s);;
t :
  'A.{metadata : metadata, track_marks : track_marks}
  where 'A is a set of internal tracks = source.tracks(source=blank)
# s' = source({audio=t.audio});;
s' :
  source(audio='A)
  .{
...
  }
  where
    'A is a track and a track of type: pcm, pcm_s16, pcm_f32, canvas, metadata or track_marks =
  <source(id=source, frame_type={audio : 'A}
where
  'A is a track of type: pcm, pcm_s16, pcm_f32, canvas, metadata or track_marks and a track>.{...}
# t;;
- : 'A.{audio : 'B, metadata : metadata, track_marks : track_marks}
    where 'A is a set of internal tracks,
      'B is a track and a track of type: pcm, pcm_s16, pcm_f32, canvas, metadata or track_marks = source.tracks(source=blank)

It's a bit hacky but it seems to work fine.