Do not retry client requests when failing with ContentTooLargeException

[swallez]

The default configuration of the Low Level Rest Client is to buffer the response body up to 100 MB. If that threshold is exceeded, a ContentTooLargeException is thrown in the lower layers of LLRC.

When such an exception occurs, LLRC will retry the request, which has some bad side effects:

• for idempotent requests like a regular search, there's no point in retrying since the response will always be too large.
• for scroll searches, retrying will move to the next scroll slice. With a cluster big enough or a number of scroll slices small enough, retrying may reach the last scroll slice, which is usually be smaller and may fit into the buffer limit. From an application perspective, the request will have succeeded with the last slice as result, despite having effectively skipped a large amount of results.

This PR fixes that, by ensuring that no retry happens if a request fails with a ContentTooLargeException.

Fixes #86041.

I added the v6.8.24 label to backport to 6.8 as this was found by a customer still using 6.8 and even if there's no additional 6.8 release, they may want to use a snapshot version with the fix.

[elasticmachine]

Pinging @elastic/clients-team (Team:Clients)

[elasticsearchmachine]

Hi @swallez, I've created a changelog YAML for you.

[nik9000]

I was going to say it'd be better to use the try in the caller, but seeing this thing, I get why you didn't.