fix(electric-db-collection): guard markReady against post-cleanup onError race

[prasadpilla]

Problem

When collection.cleanup() is called (e.g. on logout or team switch), the AbortController is signalled to stop the ShapeStream. However, there is a race condition: onError can fire after cleanup() has run but before the ShapeStream observes the abort signal.

The onError handler unconditionally calls markReady(), which throws:

CollectionStateError: Invalid collection status transition from "cleaned-up" to "ready"

This surfaces as an unhandled exception in applications and pollutes error tracking dashboards (e.g. PostHog, Sentry) with noise that has no actionable fix from the application side.

Root cause

In packages/electric-db-collection/src/electric.ts:

onError: (errorParams) => {
  markReady()  // ← unconditional; throws if collection is already cleaned-up
  ...
}

abortController is already in scope here — its abort() is already called during cleanup() — but there was no guard before calling markReady().

Fix

// Before
markReady()

// After
if (!abortController.signal.aborted) markReady()

One-liner. Zero new imports or complexity.

Test

Added regression test: "should not throw CollectionStateError when onError fires after cleanup"

The test wires a real AbortController into the mock, calls cleanup(), then fires onError and asserts no exception is thrown.

Relation to PR #1174

PR #1174 fixed the same class of abort-race issue for requestSnapshot / fetchSnapshot (in-flight network calls suppressed on abort). This PR covers the remaining onError → markReady() path that #1174 did not address.

Checklist

• I have followed the steps in the Contributing guide.
• I have tested this code locally.
• This change affects published code, and I have generated a changeset (if required).