Implemented a netvar offset cache

[sakura57]

Being a performance freak I was curious why you call NetVars::GetOffset continually at runtime as opposed to finding needed offsets during initialization then referring to those later.

I realized that GetOffset could probably be modified to cache values to get a similar performance benefit without having to rewrite much existing code.

Here are some comparisons of cached vs uncached. Each time value is for running NetVars::GetOffset("CBasePlayer", "m_vecPunchAngle") 5x in a row. The first result being slower is due to construction of the static unordered_map plus having to realloc. (could probably add a reserve() in there to optimize)

Otherwise it seems to have resulted in about a 10x speedup over the original, which is pretty decent. It might be slightly less than 10x in practice since this was just one offset in the cache, but still a pretty good improvement.

Cached offset finding: 31731 ns
Uncached offset finding: 52452 ns
Cached offset finding: 2770 ns
Uncached offset finding: 36096 ns
Cached offset finding: 2769 ns
Uncached offset finding: 32749 ns
Cached offset finding: 2734 ns
Uncached offset finding: 30897 ns
Cached offset finding: 4127 ns
Uncached offset finding: 217311 ns
Cached offset finding: 2751 ns
Uncached offset finding: 29702 ns
Cached offset finding: 2734 ns
Uncached offset finding: 29024 ns
Cached offset finding: 2766 ns
Uncached offset finding: 29133 ns
Cached offset finding: 2732 ns
Uncached offset finding: 29229 ns
Cached offset finding: 2709 ns
Uncached offset finding: 29210 ns

[aixxe]

Merged, thanks for your contribution.

I've moved a few things around in d3ed740 since I wasn't able to unload the library without any changes.