Harden JSON unmarshaling in job rescuer + job executor

[brandur]

Here, do a little more to harden unmarshaling in the job rescuer and job
executor. Previously, if a job's JSON didn't unmarshal successfully,
we'd log an error and move on, which could result in unexpected trouble.

Realistically, this would be a very rare circumstance though luckily --
the Postgres/SQLite jsonb fields guarantee the formatting of the JSON
data, so the only way for an unmarshaling problem to occur is if a job's
data was stored initially, and then the JSON Go struct later change to
be incompatible (i.e. string changed to an int or something of that
nature).

Here, during an unmarshal error, use the job's standard retry schedule
and back off, discarding the job if it's at the end of its allowed
retries. This approach is best because it gives the user a chance to
notice the failure and correct a potential unmarshaling problem by
either manipulating the job row's data or fixing their Go struct.

Fixes #1323, but also addresses a similar problem in the job executor
that Codex found while working.

[brandur]

Tool failure on this one that caused the original title to be wrong. Correct PR title/description now.