Add basic insert/delete benchmarks

[mhart]

These don’t actually benchmark MemDOWN directly, but rather compare different strategies for inserting/deleting keys. (I'm not exactly sure as to the best strategy for comparing different versions of MemDOWN - just keep a running tally of benchmark results I guess?)

I figured bench was probably a bit more lightweight than any of the other options (less deps). Running these benchmarks on my MacBook Air I get:

Scores: (bigger is better)

binary insert
Raw:
 > 0.2933411557641537
 > 0.3137747097583935
 > 0.31446540880503143
 > 0.3082614056720099
Average (mean) 0.30746066999989713

push() + sort()
Raw:
 > 0.0022455308322610924
 > 0.0023219257123087604
 > 0.0023006911276147353
 > 0.002307960386167932
Average (mean) 0.00229402701458813

Winner: binary insert
Compared with next highest (push() + sort()), it's:
99.25% faster
134.03 times as fast
2.13 order(s) of magnitude faster
A LOT FASTER

and

Scores: (bigger is better)

binary delete
Raw:
 > 0.5425935973955507
 > 0.5339028296849974
 > 0.5136106831022085
 > 0.4975124378109453
Average (mean) 0.5219048869984254

linear delete
Raw:
 > 0.15087507543753773
 > 0.14490653528474134
 > 0.14575134819997085
 > 0.14863258026159334
Average (mean) 0.1475413847959608

Winner: binary delete
Compared with next highest (linear delete), it's:
71.73% faster
3.54 times as fast
0.55 order(s) of magnitude faster
QUITE A BIT FASTER

[mhart]

I didn't want to commit this directly on master in case you guys had some comments about whether this is even useful...?

[rvagg]

lgtm except for the .npmignore, I'd rather distribute with tests and benchmarks, they don't take up enough size to warrant removal; aside from that this can be merged I reckon

[mhart]

Ah ok - that was probably my bias creeping in against anything extra going into npm due to the pain we feel in one of our apps that has a dependency tree with over 2,000 (sub)modules. Esp with us being in Aus, every little byte counts!

Have removed that npmignore commit.