Romaji to Kana Converter: Astro — Chapter 2, Exercise 2
=====================================================================

TASK
Build convertNaiveV1 (the fixed 2-then-1-character tokenizer) exactly
as this chapter describes. Run it against "kyaku" and confirm the
real "kやく" output, then run it against "kaki" (a word with no
3-character tokens anywhere in it) and confirm it produces the
correct result there -- showing the bug specifically depends on a
3-character token being present at that position, not a universal
failure.

SOLUTION

Running the naive tokenizer for real:

  console.log(convertNaiveV1('kyaku'));  // 'kやく'
  console.log(convertNaiveV1('kaki'));   // 'かき'

"kyaku" breaks exactly as the chapter describes: position 0 tries
"ky" (not a valid token) and "k" (also not valid), falls through to
emitting the raw letter "k", then correctly picks up "ya" and "ku".

"kaki" (垣, a fence, or one reading of a few other words) never runs
into the same problem, because every token it actually needs --
"ka" and "ki" -- is exactly two characters long. Since convertNaiveV1
DOES try two-character tokens (it just never tries three), a word
built entirely from two-character tokens is invisible to the bug
completely. The function isn't "broken" in some general sense; it's
specifically blind to any token that's three characters long,
wherever one happens to occur.

This is exactly why "sushi" also isn't a safe word to test this fix
against on its own -- "shi" is itself one of the base gojūon grid's
real three-character tokens (not a yōon combination at all), so it
triggers the identical class of failure convertNaiveV1('sushi')
produces the real, verified output "すsひ".

WHY THIS WORKS AS AN ANSWER
----------------------------
It confirms both real outputs against actual node execution, then
correctly narrows the bug's real scope: convertNaiveV1 doesn't fail
on "kaki" because that word happens not to contain any of the exact
tokens the bug depends on, not because the underlying flaw only shows
up on yōon combinations specifically.