This follows up on WT-16613, where we saw holding a reconfig lock for `_checkpoint_update_evict_triggers_start` and `_checkpoint_update_evict_triggers_end` led to timeout errors in WiredTiger. (See patch here).

We also cannot use compare-and-swap for doubles.

We need to redesign the configuration and access API for eviction/dirty thresholds so that we avoid holding the global reconfig lock on hot paths that read thresholds, and use a representation that is safe to update atomically, avoiding CAS on doubles.

Possible options: 

• Store thresholds as integers representing fixed-point values, e.g. int32_t threshold = 125; // 12.5% and read the integer atomically and convert it to a double.

• Reduce the scope of the global reconfig lock around threshold updates, or introduce a separate lock for threshold fields only.

• Represent thresholds as part of a config snapshot struct, atomically swap the pointer on reconfig