Investigate options and possible overhead of global string truncation

To ensure some user errors don't lead to overblown payload sizes, we want to truncate all strings in the payload to a max length, similar to what we already to with strings that are stored as keywords in Elasticsearch. This could potentially be quite overhead-heavy, so we need benchmarks of different approaches

* explicitly wrap all strings in a `maybe_truncate` function that either returns the string itself or, if necessary, a truncated version
* loop over the payload and truncate all strings before serializing
* modify `JSONSerializer` to truncate strings

Start with JSONSerializer and the payload serialization path described in the issue. Benchmark the three listed truncation approaches, including their overhead, and document the results and recommended approach; the issue does not specify the maximum length or a test target.

Written by the indexing model from the issue text.