This PR introduces TransformScheduler class as a central place for scheduling transform tasks. The outline of how this mechanism works is:

• There is a queue of transform tasks, sorted by their next_scheduled_time
• On transform _start a transform task is added to the queue
• Every schedulerFrequency (defaulting to 5s) the first task is being picked up from the queue and processed, i.e.: its listener is being triggered (indirectly triggering the indexer)
• On transform _stop a transform task is removed from the queue and no longer processed
• On transform failure (transient failure, to be precise), the task's next_scheduled_time is updated so that it is processed before it normally would. We are using exponential backoff scheme to calculate the appropriate delays
• SchedulerEngine is no longer used

Additionally, this PR implements unit tests for TransformTaskParams and TransformContext classes.

The following work items are not covered by this PR. These will be handled separately, in the subsequent PRs:

• configurable per-transform setting: numFailureRetries
• support for infinite numFailureRetries
• turning some of the irrecoverable errors into recoverable
• increase maximum possible transform frequency

Relates to #85016