Use Postgres error handling and logging

[solidsnack]

The panic! in append_change could be logged many not treated as a panic at all, for example.

[ruimarinho]

This would be ideal as otherwise it will crash postgres, right?

[solidsnack]

The particular line highlighted above -- panicing on a non-existent change type -- might be better handled with logging. I suspect it is impossible in practice -- what there is to log in WAL is INSERT, UPDATE and DELETE -- but it could happen.

Probably also want to look into the possibility of failure in the call to row_to_json. (Which would be indicated by a null result.)

[ruimarinho]

Ok - to safeguard against new types (e.g. pg 9.6 or 9.7+), which would not be a breaking change, monitoring the postgres error log by use of elog would be ideal.