Three major issues: 
* (prefix) one was that it tried to fix the Maven issue of remote repository uniqueness (globally), and that attempt was in fact wrong, fix is elsewhere. The reason it tried to do this is in fact to circumvent locking issues #1644.
* (manager) other issue was filter lifecycle, manager did it wrongly (data is _inherited on customized sessions_): the bug caused that one session was used to acquire filters and same filter got used for potentially other (maybe even reconfigured) session. This overlook forced filters to implement hoops and loops (to prevent recursion), but also prevented any third party code to have saying in filter operation.
* (both filters) make sure RemoteRepositories used as keys for prefixes are _same_ (normalized as bare)

Changes:
* fix manager to make sure filters and session are aligned
* fix/simplify prefix filter recursion prevention
* introduce public and documented way to inhibit prefix discovery

Note: as this PR now removes  the "hack" (that in fact tried to circumvent #1644) we are now back at state we were before #1575 (the "by make repositories unique" bit). As we see, this change (without locking fix) will cause Maven IT failures, as ITs are executed in parallel, and "prefix discovery" initiated over same local repository will/may cause locking conflicts and hence timeouts (result is sporadically failing ITs due locking timeouts). Hence, this fix is to be followed by fix for #1644 

Fixes #1654 
Fixes #1667