MongoDB's collHash validation scans a collection's record store and indexes but does not run session->verify, so there is currently no free way to collect on-disk size information during that scan.

This feature adds a cursor-open option, debug=(size_stats), that accumulates a per-b-tree size summary into new btree_size_* data-source statistics as the cursor traverses the tree. Accounting hooks the single per-page choke point in the tree walk, so it piggybacks on a scan the caller already performs rather than adding a pass. The results are read back with a statistics cursor.

The summary records raw constituents only: per-page-type counts and uncompressed bytes, leaf key/value bytes and counts, overflow payloads, and a leaf page-size histogram, all derived from the on-disk image. Derived figures (overhead, totals) are left to the consumer.

Scope/constraints:

• Row-store only; returns EINVAL otherwise.
• Kept deliberately separate from tree-walk statistics, which measure logical in-memory state from a dedicated walk.
• Consumer (collHash) owns correct single-pass usage to avoid double-counting.