Replace thread.sleep in sync exchange v2 with coroutine-based delay

[CDRussell]

Task/Issue URL: https://app.asana.com/1/137249556945/project/72649045549333/task/1215729137633076?focus=true

Description

Replaces two Thread.sleep(...) calls in the sync v2 retry-backoff loops with delay(...):

• ThirdPartyCredentialManager.retryingOnRateLimit (used when extending the account with a 3party credential and the server rate-limits the call)
• SyncAccountRepository.retryingOnTransientError (used during the 3party → ddg upgrade login)

Switching to delay required a suspend cascades a little to these:

• ThirdPartyCredentialManager.create()
• SyncAccountRepository.createThirdPartyCredential() and joinAccountFromThirdPartyRecoveryCode()
• RecoveryCodeProvider.createThirdPartyCredentialIfNeeded()
• A small refactor inside ExchangeV2Runner.shareRecoveryCodeAndAdvanceLocked so the recovery-code fetch+send runs inside the launched coroutine instead of synchronously before it.

Intentional behaviour changes

• SM mutex no longer held during the recovery-code fetch+send. In shareRecoveryCodeAndAdvanceLocked, the network work now runs inside the launched coroutine and outside mutex.withLock, so retry backoff doesn't block the dispatcher thread. A short-circuit at the top of the launch handles the common cancel-before-run race; any work that proceeds past it is best-effort, and any orphaned server-side credential is cleaned up by the BE's 5-min TTL.

Steps to test this PR

• Manual: sync with a 3rd party browser

UI changes

None.

---

Note

Medium Risk
Touches 3party credential minting and account-join upgrade flows plus v2 pairing host recovery-code timing; the runner mutex change can alter concurrency during pairing but is intentional to avoid blocking the SM.

Overview
Replaces Thread.sleep with delay in sync v2 retry backoff loops (ThirdPartyCredentialManager rate-limit and concurrent-modification retries, SyncAccountRepository.retryingOnTransientError during 3party→ddg upgrade). That drives a suspend cascade through 3party credential creation, joinAccountFromThirdPartyRecoveryCode, RecoveryCodeProvider.createThirdPartyCredentialIfNeeded, and the dispatcher’s v2 outcome mappers that call those APIs.

In ExchangeV2Runner.shareRecoveryCodeAndAdvanceLocked, host recovery-code provisioning and send now run inside the launched IO coroutine (with an early session == null exit) instead of synchronously before launch, so retry backoff and account/credential network work no longer run while the exchange state-machine mutex is held.

Affected unit tests are updated to use runTest for the new suspend entry points.

^{Reviewed by Cursor Bugbot for commit 68ef3a9. Configure here.}

[CDRussell]

• Replace thread.sleep in sync exchange v2 with coroutine-based delay #8986 👈 (View in Graphite)
• Sync v2: Remove direct integration with set_if_absent #9012
• Sync exchange v2: Fix password deriation mismatch #9014
• develop

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cmonfortep]

LGTM