importer: refine how we generate splits for import spans

Import processor planning contains the following TODO item that is unresolved:

> TODO(jeffswenson): this isn't complete. We don't actually want to
> generate splits for each index. What we want to do is generate splits
> for each span config produced by the table that does not coalesce. For
> example, a single RBR index would get split points between each of the
> ranges.
>
> We should also consider making bulkingest tolerate ingesting an SST
> that has data for multiple ranges. At the very least that will handle
> the case where KV decides to run splits we were not expecting.

This issue tracks discussion and resolution of that item.

Jira issue: CRDB-57830

Epic CRDB-62564