Problem

Size-stats measure uncompressed leaf images (dsk->mem_size). With compression adjust on, reconciliation budgets against maxleafpage_precomp (often ~4x maxleafpage, e.g. 128 KiB for 32 KiB collection leaves) then compresses toward maxleafpage on disk.

The leaf page-size histogram still ceilings at maxleafpage, so well-compressing collection leaves all land in the open-ended bucket and fullness vs maxleafpage looks nonsensical (~350% full). Indexes look sane because they compress less.

Change

* Add size-stats btree_size_leaf_hist_buckets and btree_size_leaf_hist_ceiling so consumers (Mongo log labels) do not hardcode histogram geometry.
* Bucket and publish the ceiling from maxleafpage_precomp (fallback maxleafpage if 0).
* Keep Mongo labeling for a follow-up SERVER ticket after this is imported.

Test

* test/suite/test_size_stats01.py: buckets == 9; with no compressor, ceiling == maxleafpage; existing underfull / mean-fill asserts still hold.