The new parameter `filterUserDuplicates` is set to true
everywhere else except in the import_synapse call. This
ensures direct duplicates from the same user will get
stored in an import from Synapse situation when migrating
to the mediarepo.

Closes #241

This should help memory some more. See https://github.com/turt2live/matrix-media-repo/issues/237

Theory is that by dumping the streams we'll relieve any bytes held on by the S3 libraries. 

Fixes https://github.com/turt2live/matrix-media-repo/issues/63

Part of https://github.com/turt2live/matrix-media-repo/issues/209

Part of https://github.com/turt2live/matrix-media-repo/issues/209

Fixes https://github.com/turt2live/matrix-media-repo/issues/187

Fixes https://github.com/turt2live/matrix-media-repo/issues/170

There's a race condition where a stream might not be available if the media is remote and being downloaded. After a couple download attempts, it fixes itself. This will acquire a media stream if one isn't provided to ensure that a stream is always available to the consumer.

Fixes https://github.com/turt2live/matrix-media-repo/issues/142

Fixes https://github.com/turt2live/matrix-media-repo/issues/136

Later support for MSC1711 will be done in https://github.com/turt2live/matrix-media-repo/issues/137

Part of https://github.com/turt2live/matrix-media-repo/issues/65

Fixes https://github.com/turt2live/matrix-media-repo/issues/20

Part of #58

Fixes #55

Alongside some other minor refactorings.

Part of #58

Downloads go through quite a few layers, and some of those layers are copy/pasted. Needs some refactoring, but it works.

Part of #58