Issue Summary

Clean eviction walks can abandon a tree after visiting relatively few pages when their candidate target is small, including when that tree holds a large share of the cache. Explore whether a larger minimum search budget can improve candidate discovery.

Context

• The current abandonment threshold scales with the target page count, and pages subsequently skipped still count toward it.
• A larger search budget can increase traversal or contention, and cache share does not imply pages are evictable. Initial synthetic and sys-perf experiments have not established a repeatable overall performance benefit.
• The threshold is not a strict limit on references traversed. This is an experimental option rather than a supported production setting.
• Related investigation: [WT-17234].

Proposed Solution

• Add a default-off, runtime-reconfigurable experimental option to scale the clean-eviction minimum search budget by the tree's estimated evictable bytes as a share of cache bytes in use.
• Preserve the existing candidate-yield ratio separately, and retain existing behavior when the option is disabled or the cache denominator is zero.
• Cover the calculation, default, connection-open configuration and runtime reconfiguration with focused tests.
• Compare the enabled and disabled behavior across relevant workloads, including walk cost, candidate yield and throughput; do not enable by default without evidence of a benefit and acceptable regressions.

Acceptance Criteria

• The experimental option is disabled by default and can be changed through connection reconfiguration.
• Disabled and non-clean eviction walks retain existing threshold and yield behavior.
• The performance trade-offs and limitations are documented for review.