Motivation
Caching cold-collection pages in the disaggregated victim block cache wastes capacity that should serve hot data. The victim cache is populated exclusively from eviction, via __evict_page_victim_cache() -> plh_cache_put(). Today that path admits cold and hot pages equally.

Approach
__evict_page_victim_cache() (src/evict/evict_page.c) is the only caller of plh_cache_put in WiredTiger, so the fix can live entirely in WiredTiger — no MongoDB/PALI change is required. Add an early return alongside the existing eligibility guards (root page, modified page, non-leaf) when the btree's storage tier is cold:

/* Pages from cold collections must never enter the victim cache. */
if (S2BT(session)->storage_tier == WT_BTREE_STORAGE_TIER_COLD)
    return;

Placing the guard early also skips the compression, checksum and byte-swap work the function performs before the cache put, so cold pages become cheaper to evict, not more expensive.

This supersedes the earlier MongoDB-side approach (propagating WT_PAGE_LOG_COLD into BlockCache::put() in pali_block_cache.cpp); handling it at the source means the flag never has to cross into PALI for this purpose.

Definition of Done

• Cold-collection pages are never inserted into the disaggregated victim block cache.
• A statistic reflects cold pages skipped on eviction.
• A test confirms a cold-tier btree's evicted page is not cached while a hot page is.