Add byte-based flush setting for ES

Currently `flush_size` is the only knob to control how much data a worker sends to ES, but if data varies in size, this can range from annoying (inconsistent heap use) to catastrophic on a low-margin system.

There should be some kind of `flush_bytes` setting as well, where each time a message is slated to be added to a pending bulk request the total bytes of the docs are considered.

Start by locating the worker's existing flush_size handling and the code that builds pending bulk requests for Elasticsearch. Trace how document sizes can be totaled when a message is added, then determine the expected interaction between flush_bytes and flush_size. Done means a configurable byte limit controls pending bulk request size without removing the existing message-count setting.

Written by the indexing model from the issue text.