Assigning to me to improve the error so that it reports the correct old size.

Assigning to @Gavin Halliday to improve the error message.

The dataset that is trying to fit into the payload field of 1 row of the index it's building here is unworkably large.
It's over 21 million rows in size, and has exceeded 2^32 bytes in size at the point of failure (not sure what final total size would have been).
NB: based on it's estimated record size, it might be as ~20GB

Even if less and it managed to succeed to build this index,
the resulting index would have been practically unusable, with single rows of ~4GB,
and with the nature of indexes being used to lookup and serialize rows remotely, it would have performed extremely inefficiently.

The error should be improved a bit, so that's clearer that huge amounts of memory are involved.
There is also an existing JIRA to complain if an ECL query tries to create indexes with huge blob fields like this: https://hpccsystems.atlassian.net/browse/HPCC-24259

@Jacob Cobbett-Smith