feat: use wordlist-4096 for denser encoding

[kklash]

Fixes #25

Summary

This PR introduces wordlist-4096, the 4096-length wordlist i spent way too much time creating over the winter holidays 😅 This wordlist replaces our use of BIP39 and provides even stronger guarantees (like a Damerau Levenshtein minimum edit distance between all words, which BIP39 lacks)

With wordlist-4096, each word encodes 12 bits to BIP39's 11. This permits denser encoding, allowing us to drop the default phrase length from 15 to 14, and encode more data in the phrase at the same time.

By expanding the version number, we will also have more room to grow in the future, especially because we are probably going to add passphrase-encrypted phrases, which will require a new version number from the get-go.

Phrase Encoding Changes

This table highlights the changes to the fields encoded in the recovery phrases.

Field	Size Before	Size After
Version Number	2 bits	4 bits
Entropy	128 bits	128 bits
Creation Offset	30 bits	31 bits
Checksum	5 bits	5 bits
--	--	--
Total	165 Bits	168 Bits
Words Needed	$\frac{165}{11}=15$	$\frac{168}{12}=14$

[veqtrus]

Do we care about similarly sounding words like abhor, abort? What about similarly looking ones like goalie, goatee?

[kklash]

It'd be great if we could maximize distance between all the words but it's a challenging task.

Once we release mnemonikey v1, I'll create a tag in the wordlist4096 repo and freeze the wordlist there. But for now, if you'd like to make any suggestions, the wordlist itself is open for suggested changes at https://github.com/kklash/wordlist4096. The encoding logic lives in mnemonikey, but the actual word list is now outside the scope of this repository.