* federation (ap, salmon)
* media (rich media, media proxy)
* upload (uploader proxy)
Each "part" will stop fighting others ones -- a huge federation outbound
could before make the media proxy fail to checkout a connection in time.
splitted media and uploaded media for the good reason than an upload
pool will have all connections to the same host (the uploader upstream).
it also has a longer default retention period for connections.
Related to #85
Everything should now be configured at runtime, with the exception of
the `Pleroma.HTML` scrubbers (the scrubbers used can be
changed at runtime, but their configuration is compile-time) because
it's building a module with a macro.
According to [the S3 docs][s3], the characters safe for use in object keys are:
* 0-9
* a-z
* A-Z
* !
* -
* _
* .
* *
* '
* (
* )
(The / character is not listed but mentioned being safe outside of the list.)
Several characters that are valid in filenames can cause problems, for example
spaces are not valid in URLs and need to be escaped,
sequences of spaces can become squeezed by S3,
some characters like \ are documented to require “significant special handling”.
To avoid these problems, this change encodes the filename
before using it as part of the S3 object name
by replacing all characters except those documented as “safe” with dashes.
[s3]: https://docs.aws.amazon.com/AmazonS3/latest/dev/UsingMetadata.html
