recomputeViews should skip only document in failure not the entire bucket

The RecomputeViewsAction is configured with fallback to continue on failure.
This means that if we cannot compute the views for a document (systematic timeout during conversion) after retries the entire record (potentially a bucket of 100 documents) is skipped.

That being said the bulk action is mostly used to process views of a single document.

We should adapt this bulk action, timing out on conversion has to be expected and only the document should be skipped, the bulk command should complete.

Also, despite the shell executor that set a timeout lower than the transaction duration, this doesn't prevent a transaction timeout with a stream failure, because there are many conversions to run, so the first is returning as expected before the tx timeout, but after the minimum timeout duration is applied (1s) so we exceed the tx duration.