Problem:

__clayered_search_near_int in src/cursor/cur_layered.c has grown to ~185 lines in a single flat body. Ingest probing, stable probing, truncate-list filtering, closest-cursor arbitration, and ingest-side tombstone skipping all share the same scope, with state threaded between them via locals.

The function has accreted over several landings (WT-16811, prepared-conflict handling, earlier features). Each change was reasonable in isolation, but the combined cognitive load is now high enough that:

• Any follow-on change has to re-read the whole body to stay safe.
• Correctness-sensitive additions (e.g. the truncate-list filter on the stable path) end up buried inside an unrelated block, which makes it easy for a future read-path feature to forget an equivalent check in the right place.
• The function is hard to add targeted assertions, verbose logging, or unit coverage to, because there is no smaller named surface to hook into.

Goal:

Reduce the cognitive load of __clayered_search_near_int. Shape is up to whoever picks this up; this is a readability refactor and must not change behavior.

Acceptance criteria:

• Top-level control flow of layered search-near is obvious at a glance.
• Existing layered cursor and fast-truncate tests continue to pass.
• No API or performance change.

Non-goals: fixing bugs, changing semantics, reworking ingest/stable arbitration.