fix: no-op on already-cached records in covert_insert_records_to_cache

[Sheikh Nasrullah (snash4)]

Problem

After loading a CPR snapshot via new_from_cpr_snapshot, inserts panic with:

internal error: entered unreachable code: Base page should not have op type: Cache
covert_insert_records_to_cache (leaf_node.rs:285)
upgrade_to_full_page
LeafEntryXLocked::insert

The CPR sweep captures PageLocation::Full pages (Cache op-types) verbatim into the BASE section of the snapshot file. After recovery those pages load as
PageLocation::Base with Cache records. The next mini-page overflow triggers
upgrade_to_full_page → covert_insert_records_to_cache, which hits unreachable!() on the existing Cache record.

Fix

Change the Cache | Phantom => unreachable!(...) arm to a no-op.
A record already in the target state needs no conversion — symmetric with
convert_cache_records_to_insert, which already tolerates the equivalent case as a no-op rather than panicking.

Why it is safe

• Symmetric: the inverse function already treats its already-converted case as a harmless no-op. This makes the forward direction consistent.
• Semantically correct: the function's job is "mark every record as cache." A record already Cache is already marked; Phantom is the cache-tier form of a delete and likewise already in the intended state.
• No data change: only op-type tagging during a tier move; keys/values are untouched.

Repro

The bug can be reproduced
https://github.com/snash4/bf-tree/tree/covert-panic-fix

[Sheikh Nasrullah (snash4)]

Already in cache state — e.g. a page restored from a CPR snapshot that was
promoted at snapshot time, then re-promoted by a write after recovery.
Nothing to convert. Mirrors convert_cache_records_to_insert, which already handles the symmetric case as a no-op.