feat(stellar): add cancellation and timeouts to the remaining Horizon helpers

[aratass]

Closes #235.

Summary

findStrictReceivePath(), buildMultisigStealthWithdraw() and submitOfflineStellarTransaction() could not be cancelled, and the account lookup and the submission had no deadline, so a stalled Horizon request could hang them. This PR brings all three in line with the Horizon and RPC clients from #202:

1. Optional signal and timeouts on each helper. Path finding and the multisig account lookup take them in their options object; submission takes a new optional third argument, submitOfflineStellarTransaction(xdr, network, { signal, timeouts }).

2. Every failure maps onto the error taxonomy.

   Failure	Error
   The caller aborts, before or during the request	New RPCAbortedError (WRAITH/NETWORK/RPC_ABORTED), the signal's reason on cause
   A timeout fires	RPCTimeoutError, attempt: 1
   Horizon answers with a non-2xx status	RPCRequestError, the Horizon body on context.responseText
   A success response that is not JSON, or an account record without its sequence, thresholds or signers	HorizonResponseError from fix(stellar): read and validate Horizon strict-receive responses #251, with the failing part on field

3. Fake-timer tests for every path of all three helpers.

Design

• Built on fix(stellar): read and validate Horizon strict-receive responses #251. Path finding keeps its request format, record validation and capped body read; this PR only links the caller's signal to its AttemptDeadline and lets timeouts override the Horizon client defaults.

• The account lookup and the submission go through one internal helper, src/chains/stellar/request.ts (requestHorizonJson), which sends a single attempt through the same AttemptDeadline. The fetch and the body read race the deadline, so a fetch that ignores the signal still times out, and timers are cleared once the request settles.

• AttemptDeadline gains abort(error). A caller abort cancels the in-flight fetch and rejects with RPCAbortedError instead of a bare AbortError. An already aborted signal rejects before anything is sent.

• No retries. These helpers made one request before and still do; a submission must not be repeated blindly.

• Defaults:

  Helper	Default timeouts
  findStrictReceivePath, buildMultisigStealthWithdraw	10 s to the headers, 30 s in total, as in the clients
  submitOfflineStellarTransaction	No headers timeout, 60 s in total

  Horizon answers POST /transactions only once the transaction is in a ledger or its own wait expires, so a short headers timeout would cut off normal submissions. The JSDoc and docs warn that a timed-out or aborted submission may still be applied, so callers should look up the hash before resubmitting.

• BuildMultisigStealthWithdrawOptions already has timeout, the transaction's validity window in seconds. Its doc now says so, to keep it apart from the new timeouts.

Behaviour changes (under Changed in the CHANGELOG)

• buildMultisigStealthWithdraw(): a non-2xx status from the account lookup is now an RPCRequestError instead of a plain Error, and an incomplete account record is a HorizonResponseError.
• submitOfflineStellarTransaction(): reports the full /transactions URL on RPCRequestError (it was the base URL), throws HorizonResponseError for a success body that is not a JSON object, and gives up after 60 s by default. { timeouts: { requestMs: 0 } } restores waiting indefinitely.

Semver

Additive: new optional signal and timeouts options, a new optional third argument on submitOfflineStellarTransaction(), and the new RPCAbortedError export. The error-type changes and the 60 s submission default above are behaviour changes, so the CHANGELOG entries sit in the Upcoming: 2.0.0 section (Added and Changed).

Bundle size

After #251, develop measures 30.22 kB of the 30.5 kB Stellar ESM budget; this PR brings it to 30.68 kB. As BUNDLE_SIZE.md asks, the budget change is in the same PR, in its own chore(size) commit, with a note in the table (new budget 31 kB). Happy to drop that commit and trim instead if you'd rather keep 30.5 kB.

Tests

• test/chains/stellar/helper-timeouts.test.ts (35 tests). Each helper runs the same nine fake-timer cases: the signal reaches fetch and no timers are left; an already aborted signal (no fetch); an abort in flight; the default timeout fires at exactly its value, not 1 ms before; custom timeouts; a fetch that ignores the abort; a body that stalls after the headers; an HTTP error with its body; a success body that is not JSON. Plus account record and submission response checks.
• test/errors.test.ts: instanceof chain, code, docs link and non-serialised cause for RPCAbortedError.

Checks run locally: pnpm format:check, pnpm build, pnpm test (1,527 SDK tests and the Svelte package), pnpm test:types, pnpm test:exports, pnpm size, pnpm api:check (reports updated in etc/) and pnpm pack:check.

Note

docs/offline-signing.md imports the offline helpers from @wraith-protocol/sdk/chains/stellar, but src/chains/stellar/index.ts does not export them. I left the exports unchanged to keep this PR to the issue. Happy to export them here or in a follow-up.

[drips-wave]

@aratass Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits

[truthixify]

Cancellation and deadlines are consistently threaded through the remaining Horizon helpers. The focused timeout/error suite passes 75/75 locally and all required CI checks are green.