feat: Move Lock-related logic to LockManager class

[vyredo]

What kind of change does this PR introduce?

This PR move logic for Lock from the GoTrueClient class by decomposing it into smaller, more focused class. Currently, GoTrueClient is a monolithic class containing several responsibilities including DOM tab visibility, lock management, session/token handling, and more. By grouping related logic into separate classes will improves code organization, maintainability, and testability.

What is the current behavior?

Changes are made without introducing big changes to the logic.
Key optimizations include eliminating redundant await operations and replacing Promise.all with Promise.allSettled to ensure all promises complete execution rather than terminating on the first error.

old logic

try {
  this.lockAcquired = true

  const result = fn()

  this.pendingInLock.push(
    (async () => {
      try {
        await result
      } catch (e: any) {
        // we just care if it finished
      }
    })()
  )

  await result

  // keep draining the queue until there's nothing to wait on
  while (this.pendingInLock.length) {
    const waitOn = [...this.pendingInLock]

    await Promise.all(waitOn)

    this.pendingInLock.splice(0, waitOn.length)
  }

new logic

try {
  this.lockAcquired = true
  const result = fn()

  // make sure fn is the last for this batch
  this.pendingInLock.push(result)
  await this._drainPendingQueue()

  // result have already completed, just unwrap the promise now.
  return await result
}

What is the new behavior?

The new implementation maintains functional equivalence with the original code while providing a cleaner architecture. The refactored code:

1. Improves separation of concerns by extracting related functionality into dedicated classes
2. Simplifies the promise handling logic and extra method for queue draining mechanism
3. Ensuring all promises complete regardless of individual failures