WL#2342: Redesign LCP

Redesign LCP according to the following:
* LCP is local (no distributed communication)

* Each node only knows what it's own replicas LCP status

* LCP starts locally when node is starting to get out of REDO log space
  or using a timer.

  TimeBetweenLCP: in seconds
  REDOpctLCPStart: in percent

* During SR, since each node only knows it's own LCP status
  they need to communicate much more to find out how to restart
  
  Exactly how this should take place is to be investigated
---

The main reason to redesign as suggested is that
LCP is by far the buggiest distributed protocol in NDB