io: public interface for io:: subsystem

[andrwng]

Today, prospective users of the io:: subsystem that want to read and write data via the pager and paging data source must interact with them directly. This is awkward for a couple reasons:

• The pager and page cache are meant to be internal implementation details
• Usage patterns expectations of the pager are forced onto callers (e.g. not reading so much that pager::read() returns a large vector), which makes it a bit clunky to use

Perhaps some of these details can be hidden away via some file-like abstraction built atop the pager and co.

JIRA Link: CORE-2015

[dotnwat]

This makes sense. In addition, what kinds of functionality will you need to get started? For example, pager lacks fsync so I think that is needed. That might be about it, right?