There can be all sorts of issues when queueing an ingest, so we'll just
log their exceptions and continue so we can get a sense of why these
documents are failing.

This was baseURL in the Pydantic model, but should be baseUrl to match
the existing Algolia schema.

Rather than use logging.exception, which places the stack trace outside
the structured log, using logging.error instead and log the exception
within the JSON structured log message.

- Format the kafka key/value data as a dict for natural JSON
  representation; the json() method was creating a string.
- Bind this context to the logger so that all log messages share it.

- Add the number of records and the surrogate key of those records so
  they can be searched later.

Configure the logger when running ook audit and other CLI apps to get
structured logging. This is useful since the CLI is primarily used for
Kubernetes Jobs.

See if this exception is now nicely formatted by structlog for us.

Turns out structlog does format the exception trace in the structured
log after all, so this is fine.