history and statistics files are sometimes corrupted because of invalid json

We do 2 things to improve this:
 1. we use json.load to read the previous file (this will detect if the file is corrupted)
 2. we use json.dump to write the entire file at once in a temporary file and then we move this file to be atomic.

Probably this will consume more resources.

See merge request nexedi/slapos.core!728