What type of PR is this?

• API-change
• BUG
• Improvement
• Documentation
• Feature
• Test and CI
• Code Refactoring

Which issue(s) this PR fixes:

Fixes #25740

What this PR does / why we need it:

NewService creates a live log store before loading local metadata and restoring persisted replicas. If either initialization step returned an error, the constructor returned without closing the store. The orphaned Dragonboat NodeHost kept its gossip/Raft listeners, TanDB handles, directory lock, and truncation worker alive, so an in-process retry with the same configuration failed with an address conflict.

This change closes the store on both error paths while preserving the original initialization error. The successful startup path, on-disk metadata, and RPC behavior are unchanged.

The regression tests exercise both sides of the failure window:

• an injected metadata open error before replica restoration;
• dragonboat.ErrShardAlreadyExist after one persisted replica has already been restored.

Each test retries construction with the same UUID, data directory, and network addresses, proving that the NodeHost, workers, listeners, and storage handles were released.

Validation (with the standard MatrixOne CGo environment):

• go test -v -count=1 -timeout 120s -run 'TestNewServiceClosesStoreOn(Metadata|ReplicaStart)Failure$' ./pkg/logservice
• go test -race -count=1 -timeout 180s -run 'TestNewServiceClosesStoreOn(Metadata|ReplicaStart)Failure$' ./pkg/logservice
• go test -v -count=1 -timeout 300s ./pkg/logservice
• go build ./pkg/logservice
• go vet ./pkg/logservice