Hash hosts

[attilaolah]

Instead of keeping a map of hosts, we can hash them and keep the hash sums to save space. Hashes are arrays, which means a fixed size, which is easier to manage and behaves more predictably wrt memory consumption.

After some profiling with the hosts file from http://hosts-file.net/download/hosts.txt, I noticed the following:

Memory	string [Mb]	[8]byte [Mb]	improvement
Alloc	107.87	60.56	43.86%
TotalAlloc	215.78	209.75	2.79%
Sys	131.24	67.86	48.29%

The runtime for parsing the entire host file did not change noticeably (it stayed around 3.4 seconds).

So far the only downside of this approach seems to be that there is no way to list blocked hosts in the web UI. Because of that, we can only use this approach for the public hosts files (which are the huge ones anyway, private hosts files will likely be much smaller).

Closes #9, for now (we could still investigate that later).

[attilaolah]

Note that I'm halving the MD5 sum here, reducing it to 8 bytes (64) bits instead of 16 bytes. This gives us a chance of collision of one in sqrt(2⁶⁴), which is roughly one in four billion (1/4294967296). A hosts file containing that many hosts would be more than a gigabyte.

[attilaolah]

Out of curiosity, I also tried using uint64 as the may key, which is also 8 bytes. It behaves exactly like [8]byte, both speed and memory wise.

I'll merge this, seems like nothing broke.