utilities/python_influxdb_downsample#9: Record failures to disk

I guess the best way to go at this is probably to have a file per write/batch.

That way, things can more easily be removed if they're clearly never going to work.

I also think this mode should be optional (and default to off) - having a YAML representation of failed points being written out sounds like a good way to fill your disk.

OK, so the dump side is largely implemented.

New config is supported

failures:
    record: True
    path: dump.d

if record is True, then when a write failure occurs the payload will be written out into path (dump.d in the above example).

The dump's filename is of the format f"{op_influx_id}-{bucket_slug}-{dt.strftime('%Y%m%d%H%M%S')}.yaml.gz"

The structure inside the file is as follows

{
    "output_id" : op_influx_id,
    "bucket" : bucket,
    "records" : new_records,
    "failure_time" : dt
}

What we don't currently have is support for reading that data in and replaying it.

I think that can come later though - it's probably better implemented as a standalone(ish) utility.