Add benchmark for undo_index.

[greg7mdp]

This PR adds a benchmark directory containing a benchmark designed to exercise the undo_index specifically.
I ran it 9 times with the older version of chainbase (before the memory optimization) and identically 9 times with the newer version (using 42 bit offsets as pointers for memory efficiency).

The memory optimized version does not appear to be significantly slower (test with gcc_12, release build, AMD 7950x). These numbers show an average 3.05% slowdown with the memory optimization with intensive chainbase use, so I would not expect to see a noticable slowdown in leap.

Before memory opt
Bench time	13.39s
Bench time	12.42s
Bench time	12.58s
Bench time	12.36s
Bench time	13.49s
Bench time	12.69s
Bench time	12.41s
Bench time	12.33s
Bench time	12.54s

After memory opt
Bench time	12.93s
Bench time	12.66s
Bench time	12.91s
Bench time	12.76s
Bench time	13.13s
Bench time	12.56s
Bench time	12.54s
Bench time	13.34s
Bench time	12.56s