Use PerBufferUnread by default for backlog fetching rather than
PerBufferFixed. In cases with many slow channels or multiple PMs,
this reduces the amount of backlog fetched on connect by ignoring
content already read. This also increases the chances of seeing
highlights past the previous fixed limit of 500 lines.
This modifies new installs and existing configurations if the prior
default value wasn't changed.
In most cases, this should speed up connecting. However, if
problematic, the version upgrade system should be used to preserve
old defaults for existing installs.
Closes GH-247.