fix: reject word lists without exactly 2048 words instead of emitting broken checksums

[Kshot3000]

Problem

addressToChecksum (JS) and address_to_checksum (Rust) map 11-bit indices (0–2047) into whatever word list the caller passes, without checking its length:

• JS: a truncated list silently produces a broken checkphrase. With the bundled list cut to 100 words, the Satoshi test address returns ["ahead", undefined, undefined, undefined, undefined] — an anti-poisoning checksum with four of its five words missing, and no error anywhere. An empty list returns five undefineds.
• Rust: the same input panics with index out of bounds: the len is 100 but the index is 1016, which never tells the caller what the actual contract is.

The Dart implementation already validates the length in its constructor (ArgumentError: Word list must contain exactly 2048 words), and both the Rust and JS loaders validate the lists they ship — but the checksum functions themselves accept any list, so any caller passing a custom or accidentally truncated list hits this.

Fix

• JS: addressToChecksum throws Word list must contain exactly 2048 words, found N before deriving.
• Rust: address_to_checksum asserts the same condition with the same wording load_word_list already uses for its error, so the panic names the contract instead of an index.

No behavior change for valid 2048-word lists.

Tests

• Jest: new case covering 100-word, empty, and 2049-word lists (all reject with the message above). Full suite 6/6, including all 1,171 cross-platform vectors.
• Rust: two should_panic cases pinning the message for short and empty lists. cargo test --release 6/6; cargo fmt --check clean.

Reproduced before the fix with a scratch spec (JS result above) and a scratch Rust test (panic message above); both now fail loudly and clearly instead.

---

Submitted by @Kshot9000 (https://x.com/kshot9000) — independent builder in the Quantus ecosystem. Tips welcome in QTC: qznY8nwuvWcCCVys4da1oQdysyh8YUZYRjRqgk3S8Wos8kbau

cc @Quantus-Network

[n13]

Reviewer model: GPT-6.1 Sol

APPROVE — no blocking findings at 0a6191fbe609a23bdcae08d01dc6bc85d7cc9abc.

The Rust and JavaScript guards enforce the existing 2048-word contract before PBKDF2 runs. Valid-list derivation and public signatures are unchanged. The regression tests verify the explicit failure messages; both implementations still pass all 1,171 shared checksum vectors.

Validation passed in a detached worktree:

• cargo test --locked --release — 6 tests passed, including the shared vectors.
• npm ci --no-audit --no-fund — dependency installation and TypeScript build passed.
• npm test -- --runInBand — 6 tests passed, including the shared vectors and short/empty/oversized list rejection.
• cargo clippy --locked --all-targets --all-features -- -D warnings.
• cargo +nightly fmt --all -- --check.
• python3 scripts/sync_wordlists.py --check and git diff --check.

The latest GitHub CI snapshot has passing format, wordlist sync, JavaScript, Linux Rust, and Clippy/doc checks; macOS/Windows Rust and Dart jobs are still running.