new method and improvements

[ebrensi]

Hello again and happy new year. In this pull request I have

• Changed all var statements to the modern let and const where appropriate, in the module itself and the tests and benchmarks.

• Added a FastBitSet.fromWords constructor, for creating a new FastBitSet if we already have words. Actually this is meant more for the TypedArray version, which I will make a pull request for soon.

• Added a (sort of) new method called difference2. The difference method (AND NOT) is not commutative like the other operations, so the existing in-place operation may modify the wrong object for a potential use. In my case, I am monitoring new changes from one moment to the next, so I want to modify the old FastBitSet, not the new one. For FastBitSet objects A and B, A.difference2(B) performs the same operation as A.difference(B), in-place, except it modifies B instead of A. I made it into a new method instead of modifying the existing one because I did not want to affect performance of the original one, which is faster. difference2 is slower than difference, but it is still much faster than new_difference.

• Added benchmarks for the change and new_change (XOR) methods that I introduced in my last pull request.

• After running the benchmarks I discovered that change and especially new_change were too slow, so I made them both more efficient.

• Added all the required libraries for Benchmarking into package.json dev-dependencies, and a new npm run benchmark script, making it easy to run benchmarks.

Note: I disabled the infusion.BitSet.js benchmark for ArrayBench() because for some reason it was making my laptop hang.

[ebrensi]

Here are relevant benchmarks from my laptop

change

starting inplace change (XOR) benchmark
FastBitSet (inplace) x 90.15 ops/sec ±0.31% (79 runs sampled)
infusion.BitSet.js (inplace) x 0.79 ops/sec ±2.85% (6 runs sampled)
tdegrunt.BitSet (inplace) x 0.78 ops/sec ±4.55% (6 runs sampled)
roaring x 432 ops/sec ±1.14% (97 runs sampled)
Set (inplace) x 4.23 ops/sec ±3.03% (15 runs sampled)

new_change

starting change (XOR) query benchmark
roaring x 1,064 ops/sec ±39.31% (42 runs sampled)
FastBitSet (creates new bitset) x 26.86 ops/sec ±3.93% (50 runs sampled)
mattkrick.fast-bitset (creates new bitset) x 10.18 ops/sec ±7.47% (30 runs sampled)
Set x 1.99 ops/sec ±4.48% (10 runs sampled)

difference2

starting inplace difference benchmark
FastBitSet (inplace) x 7,865 ops/sec ±8.90% (100 runs sampled)
FastBitSet (inplace2) x 7,274 ops/sec ±7.82% (99 runs sampled)
TypedFastBitSet (inplace) x 6,503 ops/sec ±2.89% (97 runs sampled)
infusion.BitSet.js (inplace) x 19.36 ops/sec ±8.89% (37 runs sampled)
tdegrunt.BitSet (inplace) x 22.18 ops/sec ±5.03% (40 runs sampled)
roaring x 50,038 ops/sec ±1.19% (94 runs sampled)
Set (inplace) x 4.42 ops/sec ±2.36% (16 runs sampled)

[lemire]

Looks good. Please consider my specific comment above.