TBS: processor fails and stops when transaction/span is too large

[simitt]

APM Server version (apm-server version):
8.10 (and previous)

Description of the problem including expected versus actual behavior:
Under specific circumstances, the apm-server tail sampler process returns and stops without shutting down the apm-server. This seems to happen in relation to tail based sampling when a span cannot be deleted from the local storage due to its size (https://github.com/elastic/apm-server/blob/main/x-pack/apm-server/sampling/processor.go#L536).
This manifests in an error message such as failed to delete span from local storage: Txn is too big to fit into one request.
Right now the only way to mitigate is to manually restart the apm-server.

To fix this we need to

• already reject events that exceed a max size when they are received, as well as ensuring that events of allowed size can be processed by the system
• avoid situations where the processor stops running without stopping the apm-server.