The persisted state is mostly votes, and it holds their content several
times over: every vote carries the proposal value, the bundle that
proved a threshold repeats the signatures of the votes that formed it,
and msgpack writes every field name. zstd at level 3 removes the
repeats and keeps the signatures and proofs: a mainnet snapshot shrinks
from 433 KB to 105 KB.

This release restores a compressed snapshot, recognized by its frame
magic, and still persists the uncompressed form; the next release
persists the compressed form, so that the release before this one,
which does not decompress, restores what this one persists.

Ports the Falcon-1024 scheme support onto the refactored layout:
isEd25519Scheme and the scheme-aware signer encoding in arc60Wire,
SIGN_DATA_VERIFICATION_UNSUPPORTED in signArc60Data, the scheme field
on the models, and the version bump to 1.9.0. Its tests move into the
corresponding *.spec.ts files.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUJASfdAAc5HNUZoEkPjAd

e2e/ bundles dist/ into a small Vite dApp, the way an integrator's
bundler would, and drives it in Chromium with all external traffic
stubbed or dropped: the modal elements and legacy globals register, the
version shows in the header, the rendered QR decodes back to the
WalletConnect pairing URI, closing the modal rejects
CONNECT_MODAL_CLOSED, and the extension connect runs inside a real user
gesture (the stub refuses without navigator.userActivation).

Runs as its own CI job (pnpm test:e2e) and uploads traces on failure.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUJASfdAAc5HNUZoEkPjAd

src/__integration__ drives the SDK only through its entry point, with
the real modal elements, storage and algosdk keys, and fakes only the
outside world: a WalletConnect v1 client modelled on the real one, a
Pera Mobile that scans the rendered QR and signs with a real key, a
window.pera extension that only connects inside a user click, and the
config/algod endpoints. Covers connect (QR, extension, modal closed),
reconnect after reload, signing (toast, redirect modal, transaction
match, TransactionSigner), signData/ARC-60 verification, wallet-side
disconnect, empty signatures and disconnect.

Vitest now runs unit and integration as two projects
(pnpm test:unit / test:integration); coverage over both is gated at
95% lines/functions/statements and 90% branches. Known-wrong behaviour
found on the way is recorded as it.todo.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUJASfdAAc5HNUZoEkPjAd

Behavioural unit tests for every modal custom element (attribute wiring,
conditional views, the extension option event, close and download
actions) and for the gaps in util/ and transport/: the Pera Web
new-tab flows, the webview bridge, deep links, the signed-transaction
match check, ARC-60 signer resolution and the extension provider.
Third-party renderers (qr-code-styling, Lottie) are faked at the module
boundary; everything else runs for real.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AUJASfdAAc5HNUZoEkPjAd