Make the Kirat Rai ligature states ligature-transparent

[youdie006]

The crate measures the same Kirat Rai ligature two ways:

"\u{16D68}".width()                  // 1  precomposed
"\u{16D63}\u{16D67}".width()         // 1  decomposed
"\u{16D63}\u{200D}\u{16D67}".width() // 2  decomposed + ZWJ

KIRAT_RAI_VOWEL_SIGN_E and _AI (src/gen/width_info.rs:39-40) are the only two of eight script-ligature states carrying none of the ligature-transparency bits, so is_ligature_transparent is false and src/lookup.rs:149-155 never skips the ZWJ. scripts/unicode.py:214-215 states the rule they break: "5th from top: whether this is unaffected by ligature-transparent (if set, should also set 3rd and 4th)". #53 established the convention for the first six states; #74 added Kirat Rai without it.

The new constants use the same 0b0011_1100_ prefix as LISU_TONE_LETTER_MYA_NA_JEU and KHMER_COENG_ELIGIBLE_LETTER. The edit is in scripts/unicode.py; src/gen/width_info.rs is regenerated output.

Scope and verification

Differential over every Unicode scalar value as the inserted character, six <base><mid><closer> sequences, pristine vs fixed, both width and width_cjk:

sequence	cells changed	transitions
Kirat Rai 16D63..16D67	528	all 2 -> 1
Kirat Rai 16D63..16D68	528	all 2 -> 1
Hebrew / Lisu / Khmer / emoji ZWJ	0

row	result
new test on master	FAILED, "𖵣\u{200d}𖵧" has the wrong width, left 2 right 1
with the fix	ok, 37 passed
revert	FAILED
transparent bit without the ZWJ bit	FAILED
fix _E only	FAILED

CI rows run locally: cargo test, --no-default-features, cargo fmt --check, clippy both feature sets under -D warnings, cargo doc under -D warnings, cargo +1.66.0 check --lib --features core,std — all exit 0. The regen gate (rm tests/emoji-test.txt && cd scripts && python3 unicode.py, then git update-index --refresh && git diff-index --quiet HEAD --) exits 0 on the committed tree, and all five src/gen/*.rs md5s are unchanged by that regeneration.

One thing I did not check: the Unicode core spec text on whether Kirat Rai vowel-sign ligatures must be ligature-transparent. I verified only that this crate is inconsistent with itself, with its generator's stated bit contract, and with its own docs. If the intended answer is the other direction, src/lib.rs:80 is what needs changing and I will send that instead.

---

Disclosure: I used Claude (an AI assistant) while preparing this change. Every result above I ran and verified myself.

[Jules-Bertholet]

Thanks for the report!

I am inclined to say we should resolve this by changing the documentation. The Kirat Rai rules are intended only to ensure canonically equivalent strings have the same width, nothing beyond that. Unfortunately there is no Unicode spec for this, and no Kirat Rai monospace fonts.

(As an aside, the excessively verbose Claude output is quite grating to read in PR descriptions, and wastes everyone's time. Please try to cut it down in the future)