RFE: speed up resolving syscalls by name by using hash map

[varqox]

It slightly mitigates #153.

[coveralls]

Coverage decreased (-2.7%) to 89.358% when pulling a272efe on varqox:master into 3570b5c on seccomp:master.

[pcmoore]

I'm not entirely sure how I feel about this approach at the moment.

If I understand you correctly @varqox, the main motivation for this was issue #153, which I believe we've addressed in the master branch (scheduled for the v2.5 release). How do you feel about this PR now?

[varqox]

Yes, I did this because before the fix from #180 resolving the name of a syscall was the primary thing that libseccomp did (based on perf analysis). Now it is only roughly 10% of what libseccomp is doing.
Moreover, my solution reduces name resolving only by about 50%. Also, it is not a clean implementation because of the lack of compile-time data manipulation (building the hash map) in C.

For now, I don't think that this solution is worth ~5% performance increase of libseccomp.

[pcmoore]

For now, I don't think that this solution is worth ~5% performance increase of libseccomp.

That is my thinking at the moment too. While there are definitely improvements to be had with respect to syscall name/number resolution, I'm not sure this is the right fix (although a hash map is likely a good idea).

With the conversation above in mind, I'm going to close this PR. Thanks again for your patience and help in resolving the #153.