Revisited how transitions are handled by the cmx 3600 adapter.

[gplsteph]

The code handling transitions in the cmx 3600 adapter was hard to follow and was introducing problems. These changes
fix these problems and make the code easier to maintain.

• Transitions lines are now collected when parsing EDL lines and treated by the ClipHandler.
• The Clip source and record points are replaced with the transition values when the line is parsed.
• The transition in_offset is now 0 and the out_offset is the transition duration, matching in the timeline what would be done if the editing was read from another format.
• The transition is added just before the clip when it is added to a track, and that's it.
• Adjusted tests for the new way of handling transitions, added some other checks.
• Added an EDL read/write/read roundtrip to check that the reader and writer are compatible.
• Removed code which was not needed anymore.

Fixes #1328
Fixes #912
Fixes #977

Now the transitions read from an EDL appear at the right time in the timeline, and the timeline has the right overall length.

Associated tests.
These changes are covered by the test_cmx_3600_adapter tests. A read/write/read roundtrip test was added to ensure they are compatible with the writer.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: gplsteph / name: Stéphane Deverly (23dfc87)

[reinecke]

This is a clever way to solve the awkward parsing problem CMX transitions pose!
My only reservation is that the result doesn't exactly look like a user would expect it to, but the parsing is much more robust in this case so I think it's a good trade-off. As long as consumers are properly handling "imbalanced" transition overlaps, I don't think this should cause an issue.
Thanks for the contribution!