[fix][ml] Avoid blocking the factory-shutdown thread on serial managed-ledger close

[merlimat]

Motivation

ManagedLedgerFactoryImpl.shutdownAsync first initiates an async close (asyncClose) for every managed ledger known at the start of shutdown. A second pass, inside bookkeeperFuture.thenRun(...), then closes any ledgers that were added after that snapshot — but it did so with a serial, blocking managedLedger.close() (which waits on a CountDownLatch), blocking the thread that completes the shutdown future.

Modifications

Change that second pass from thenRun to thenCompose, and replace the blocking close() with asyncClose(), collecting the per-ledger futures and awaiting them with Futures.waitForAll. In the common case this pass is empty (the first pass has already drained the ledgers map), so it is a no-op; when ledgers were added concurrently during shutdown they now close in parallel without blocking the continuation thread. The first pass and the rest of the shutdown sequence are unchanged.

Verifying this change

Covered by ManagedLedgerFactoryShutdownTest.openEncounteredShutdown (shutdown while a ledger open is in progress) — passes.

Does this pull request potentially affect one of the following parts:

If the box was checked, please highlight the changes

• Dependencies (add or upgrade a dependency)
• The public API
• The schema
• The default values of configurations
• The threading model
• The binary protocol
• The REST endpoints
• The admin CLI options
• The metrics
• Anything that affects deployment

[void-ptr974]

Previously this path used managedLedger.close(), which bounded the wait with AsyncOperationTimeoutSeconds before continuing shutdown. With the new asyncClose() + waitForAll flow, do we still want to keep a bounded wait here?