RWI reading after instance crash takes ages

[okybaca]

After an instance crash (power-out, ...), fresh start takes very long time
because of HeapReader reading RWI database and rebuilding indexes (~half a
day in case of 300GB RWI database) resulting in prolonged instance
down-time.

I'm not sure about how RWI database works, prove me wrong please, but from
what I observe:

• all the segments indices are kept in memory during run
  (which can result in memory exhaustion as described in RWIs fill out the whole memory space for YaCy #731)

• during correct shutdown, indices are written to (.idx and .gap?) files and
  re-start is relatively fast

• if the instance crashes, no .idx & .gap files are written

Possible solution could be:

• keep the .idx & .gap files until the segment is changed

• maybe time-to-time, write .idx & .gap file, when the segment wasn't
  changed for a long time

• if running into low-memory condition, unmount (?) the oldest segments.
  check this also on start-up, because with a RWI database bigger than
  available RAM, the instance wouldn't start, resulting in memory error

• speed-up the RWI transfer to other instances, as suggested in RWI: indexDistribution.minChunkSize does nothing #724.
  RWI database on a host only grows even by moderate crawling (~200ppm,
  DHT-IN switched off) and never actually shrinks, resulting in many
  instances jam.