I was wanted to understand how many pages need to be read to find a single record in a relatively large tree. I could not find existing statistics that allowed me to get that information.

I ended up with a change that has a combination of tree walk statistics that give better insight into the size/shape of the tree and session statistics that differentiate the type of pages read by a transaction.

I'm not sure if it's worth fixing the changes up and getting them merged, but saving here in case it is.

I tested this using a `wtperf` workload which is described as:

--- a/bench/wtperf/runners/parallel-pop-btree.wtperf
+++ b/bench/wtperf/runners/parallel-pop-btree.wtperf
@@ -1,9 +1,10 @@
 # wtperf options file: Run populate thread multi-threaded and with groups
 # of operations in each transaction.
-conn_config="cache_size=200MB"
+conn_config="cache_size=2000MB"
 table_config="type=file"
 transaction_config="isolation=snapshot"
 icount=10000000
 report_interval=5
 populate_ops_per_txn=100
-populate_threads=5
+populate_threads=8
+value_sz=1024