`tokenize()` tested `hasPrefix("true")` and `hasPrefix("false")` ahead of everything
else, so any word starting with those letters was cut in two: `truncate()` came out as
`boolean(true)` followed by `identifier("ncate")`. No function is named that way today,
but the trap was set.

`parseIdentifier()` now reads the whole word and only an exact `true` or `false` becomes
a boolean; anything else stays one identifier. `parseBoolean()` is gone with it.