refactor(topic): make writer state and task ownership explicit

[aral01]

Pull request type

Please check the type of change your PR introduces:

• Bugfix
• Feature
• Code style update (formatting, renaming)
• Refactoring (no functional changes, no api changes)
• Build related changes
• Documentation content changes
• Other (please describe):

What is the current behavior?

• Logical state is distributed across the queue, reconnector status watch channel, fatal-error channel, current stream writer, and reconnect state machine.
• Queue uses tokio::sync::Mutex, although only sync work is done under mutex.
• Reconnector manually implements retry machinery across multiple state-machine functions.
• Connection task ownership and failure handling are distributed across several abstractions.

Issue Number: related #600

What is the new behavior?

• Shared logical writer state is consolidated in WriterState
• Each connection is represented by five fail-only background jobs stored in JoinSet<YdbResult>.
• Reconnector becomes straight loop: acquire initial StreamWriter, watch all spawned tasks, on fail: cancel others, in retry loop spawn new tasks. State machine and helpers are deleted.
• The standard RetrySettings API is used to retry all operations

Other information