Replace gob encoding with own writing to hash.

[s-l-teichmann]

As we already do a recursive re-coding to make gob usable instead
we can directly write to the hash during the recursion.

This makes a difference. On my machine:

main branch using gob:

> /usr/bin/time -f '%E' ./fakedoc \
   -o large.json -size 0.001  -l ../../limits.json \
   -seed pcg:81c43e3f5d1b81f2:8cc495feb113a405
0:29.03

This branch using direct hashing:

> /usr/bin/time -f '%E' ./fakedoc \
   -o large.json -size 0.001  -l ../../limits.json \
   -seed pcg:81c43e3f5d1b81f2:8cc495feb113a405
0:22.21

[bernhardreiter]

addresses #18 , in one case from 8 minutes to 30 seconds

[koplas]

LGTM