Opened a layered cursor currently takes the schema lock. 

I assume this is needed to ensure consistency across the layered table's constituent tables. 

We should investigate whether it is possible to do this locking/coordination at the level of the layered table's dhandle rather than requiring a global lock. 

Currently this is especially painful since we don't cache layered cursors, leading to benchmarks showing 50,000+ schema lock acquisitions per second. But even when we get the cursor caching properly implemented (WT-14542), it would be better to avoid a global lock if we don't need it.