scure/base is slow

[ChALkeR]

As in, 5-30x slower than a somewhat optimized impl

Lowest hanging fruits:

• Replace Array with Uint8Array for intermediate representation.
  This will also fix issues like max Array length in Chrome limiting this lib to about 60 MiB
  See feat!: use Uint8Array in alphabet() and radix #50 and feat!: unite alphabet, join, padding #51
• Use TextEncoder where available instead of .charCodeAt, which will also simplify validation
• Use TextDecoder where available to convert from uint8arr of codepoints instead of string concatenation
  See perf: use TextDecoder for building strings when available #52
• perf: trust Uint8Array.fromBase64 to check non-whitespace chars #47
• ...

Here are benchmark results
Taken on an M3 on random ~5 KiB Uint8Arrays (except for base58/base58check, which are 20-40 bytes)

Note: Firefox native codepath being slower than js is reported

[paulmillr]

It would be great to improve it, while keeping in mind one important thing:

We always prioritize bundle size over speed. Speed seems to be "good enough" for most applications, it doesn't really matter.

If it's e.g. 30 lines instead of 10 but much faster - then it's a great compromise. If it's 300 lines, then it's bad.

[ChALkeR]

@paulmillr it's mostly the other way around, this is slow because it's overcomplicated with transforms and keeping track of arrays

I'll file a PR to demonstrate

[ChALkeR]

See #51 for a demo