Faster `toString('hex')` implementation

[jawj]

In a similar vein to #245, I have an implementation of toString('hex') that's between 2x and 4x faster than the existing one, depending on the browser/engine.

See: https://jsbench.me/evm3ejel2i/3

Possible drawbacks of this implementation:

• Requires TextDecoder (does this package support IE currently?)
• Likely uses more memory, since it allocates a Uint16Array to hold intermediate data

Would you accept a PR to switch to this implementation?

[jawj]

Thinking about it, we can probably limit the extra memory needed by doing it in chunks ...

[jawj]

I've packaged this code up on npm: https://www.npmjs.com/package/hextreme. For multi-megabyte arrays it can be many times faster than the approach currently taken here.

Are you willing to consider a PR to use this package for .toString('hex')?

[dcousens]

I think we reasonably track Node LTS for the Buffer API, and with IE officially out of support - I'm happy to say we that we could drop support for that in respect of code used.

[jawj]

Thanks. I actually added a fallback implementation where there's no TextDecoder anyway, which is still slightly faster than the current approach because it does a bit of loop unrolling.

I also limited the biggest possible temporary buffer allocation to 1MB, by doing the conversion in chunks, which makes very little difference in speed.

I'll go ahead and submit a PR in the next few days then.

[jawj]

I've actually just applied the same technique to base64 encoding, so I might submit another PR for that at some point.