Optimize CFG construction

[sim642]

In light of #603, I spent a bunch of time myself trying to understand what's going on and finding ways to optimize the situation. These changes have been cherry-picked from #609 as they should be in working condition and mergeable.

Changes

1. Add timing to CFG construction with minimal indent changes.
2. Output CFG hashtable statistics.
3. Replace add and find_all with replace and the values being lists. This gives equivalent multimap behavior but makes it more explicit and notably keeps the hashtable bucket sizes much smaller. For example, on a subset of 50 OpenSSL files, instead of maximum bucket length of 198, it now is 9.
4. Decrease initial sizes of some hashtables. Otherwise there are large amounts of hashtables that have very low load factor, so they allocate unnecessarily long arrays for a handful of elements. This is also detrimental to memory usage and caching.

TODO

• Decrease initial sizes of more hashtables.
• Change multimap presentation in SCCs as well.