sql/stats: support arbitrary partial statistics over multiple spans

We disabled creating statistics with a `WHERE` clause requiring scanning more than one span in #153419. The merging logic assumes that the histogram is continuous and represents a scan over its full range, which wouldn't be the case if the stat collection was over multiple spans.

Note that partial stats `USING EXTREMES` scans two spans which are merged properly as a special case, by splitting the buckets before and after the full stat's lower bound and merging them separately. We could do something similar for partial stats with a `WHERE` clause to support those over multiple spans.

Jira issue: CRDB-54807

Epic CRDB-55121