fix(perps-controller): keep watchlist stars across overlapping hydrations and failed writes

[abretonc7s]

Explanation

This is a follow-up to #10665. Two existing races could lose a watchlist star.

• Overlapping hydrations: #performInitialization replaced #ausQueue with the AUS hydration instead of chaining onto it. Here is how a star got lost with full AUS:
  1. The user stars a market while hydration is in flight.
  2. toggleTestnet() or switchProvider() starts a second hydration at once. Its read doesn't include the star, because the star hasn't been written yet.
  3. That read overwrites local state, and the star's queued write then saves the list without it.
• Failed write: a rejected PUT restored the list from before the toggle. That discarded markets hydrated since then and later toggles, so a later star could be lost locally and remotely.

The fix:

• Hydration now runs through #ausQueue, so its read waits for earlier writes. Its edit list is registered when it is queued, so a toggle made while it waits is replayed onto the remote list too.
• A failed write undoes only its own toggle. A removed market goes back to its position. The undo is skipped when a later toggle of the same market is pending, because that toggle's own sync decides the market's state.

No in-flight hydration can still hold the failed edit. Any hydration that recorded it was queued before the edit's write, so it has finished by the time the write fails.

With only the AUS read delegated, a hydration queued after a star still replaces it, as #10665 documents. A test pins this. Trade-off: re-initialization no longer replaces the AUS queue, so a hydration read that never settles now also holds later watchlist writes. Stars still apply locally, and init stays non-blocking.

References

• Follow-up to fix(perps-controller): degrade quietly when optional messenger actions are missing #10665.
• Follow-ups: bound queued AUS hydration with a deadline that also stops a late read from hydrating; when the latest toggle of a market fails after earlier toggles of it, set the market from what AUS holds rather than undoing only the latest; an earlier successful whole-list write may already hold a later edit (pre-existing).

Checklist

• I've updated the test suite for new or updated code as appropriate
• I've updated documentation (JSDoc, Markdown, etc.) for new or updated code as appropriate
• I've communicated my changes to consumers by updating changelogs for packages I've changed
• I've introduced breaking changes in this PR and have prepared draft pull requests for clients and consumer packages to resolve them

---

Note

Medium Risk
Changes serialized AUS watchlist sync and optimistic rollback in user-facing favorites; wrong ordering could still lose stars or block writes if hydration hangs, but scope is isolated to perps watchlist persistence.

Overview
Fixes watchlist favorite races when syncing with Authenticated User Storage (AUS), following #10665.

Overlapping hydrations: Re-init (toggleTestnet, switchProvider) no longer replaces #ausQueue with a new hydration. Hydration is chained so its read runs after pending toggle writes, and toggles made while hydration is queued are recorded and replayed on the remote list (not only in-flight reads).

Failed AUS writes: On PUT failure, the controller reverts only the failed toggle (re-insert removed symbols at their prior index) instead of restoring a pre-toggle snapshot, so hydrated markets and other toggles are not wiped. A per-network:symbol pending toggle counter skips rollback when a later toggle of the same market superseded the failed one.

Changelog and state tests (stateful AUS store) document the trade-off: a hydration read that never settles can block later writes on the queue.

^{Reviewed by Cursor Bugbot for commit 5431555. Bugbot is set up for automated code reviews on this repo. Configure here.}