Alex, can you please clean up __wt_btree_get_memsize()?

I'm sure it's for historic reasons, but it's pretty strange at this point:

1. It only has one caller (__clsm_open_cursors()) that ignores any error return (might as well return a void?)
2. It sets WT_BTREE_NO_EVICTION which is weird, no-eviction has nothing to do with its primary mission in life (why is the no-eviction flag set here and not somewhere in LSM?)
3. I don't think it's possible for "child" to be NULL inside __wt_btree_get_memsize, that is, if the root page has a non-zero entries, there better be a child reference.

I'd do it myself, but I'm not sure if the no-eviction flag belongs here or should move somewhere else.