feat: non-ASCII support and optimization in murmurhash3

[dj-stormtrooper]

I added proper support for non-ASCII symbols for murmurhash3_128_x64. Previous implementation wasn't compatible with implementation in other languages (e.g. Golang, C++) for symbols with charcodes outside of 0 - 0xff range. Also it was limited in terms of uniqueness, e.g. hash('Hello, world, hi') was equal to hash('ňťŬŬůĬĠŷůŲŬŤĬĠŨũ').
Current new implementation are free of these issues

Also I refactored internal functions to achieve minimal minimal memory allocations and reduce GC load, it gained performance boost:

• Chrome: 27%
• Webkit: 58%
• Firefox: 45%

Optimizations:

• the number of new object creation is minimized
• all int64 values (int32 tuples) are passed to mutable functions, and the result is written back to the value
• int64 constants are defined outside the scope if possible

Benchmark (sorry for the minified code in the benchmark, jsbench allows to save suite only if fields are below 2048 symbols, so I tried to make it more compact)

JSBench screenshots

Webkit

Chrome

Firefox

Other implementations that I've tried:

• BigInt instead of int32 tuples feat: murmur3hash BigInt implementation #933
• Iterating through code points instead of converting to byte-array beforehand feat: murmur3hash with iteration through code points #936

[Finesse]

Please list shortly the optimizations that you applied in the PR description

[dj-stormtrooper]

Please list shortly the optimizations that you applied in the PR description

Done