Fix device plugin grpc properly stop and restart after kubelet restart

[peachest]

What type of PR is this?

/kind bug

What this PR does / why we need it:

Fix three defects in PluginServer gRPC restart logic, ensuring Stop+Start cycles do not cause panics, data races, or goroutine leaks.

Defect fixes

Defect 1: Start() reuses the grpcServer created in the constructor. On the second Start() call, serve() triggers the gRPC fatal error RegisterService after Serve (calls os.Exit(1)).

Defect 2: Stop() does not wait for background goroutines to exit — it calls close(stopCh) and returns immediately. When Start() creates new goroutines that read the new stopCh, stale goroutines may still hold references to the old stopCh, causing a data race.

Defect 3: Stop() does not clean up the Unix socket file, so net.Listen may fail on restart.

Fix summary

File	Changes
server.go	1. Rebuild ps.grpcServer = grpc.NewServer() in Start() (Defect 1); 2. Add wg sync.WaitGroup to track all goroutines (Defect 2); 3. Stop(): close(stopCh) → grpcServer.Stop() → wg.Wait() → os.Remove(socket) (Defect 2, 3); 4. All three goroutines call wg.Add(1)/Done(); 5. Add private test hook fields (dialFunc, registerKubeletFunc, prepareHostResourcesFunc) for test isolation;
register.go	Add wg.Add(1)/Done() to watchAndRegister; add registerKubeletFunc and dialFunc hook checks;

Which issue(s) this PR fixes:

Fixes Project-HAMi/HAMi#1911

Special notes for your reviewer:

• All five restart tests correctly detect the original three defects (verified by reverting the fix and confirming test failures).
• When run under go test -race, the five restart tests may still report flaky false positives due to a Go 1.26 SSA optimization that replaces function parameter reads with direct struct field reads in go statement argument evaluation. See race-detector-false-positive.md for a full root-cause analysis.

Does this PR introduce a user-facing change?:

No

[hami-robot]

@peachest: The label(s) kind/bug cannot be applied, because the repository doesn't have them.

Details

In response to this:

What type of PR is this?

/kind bug

What this PR does / why we need it:

Fix three defects in PluginServer gRPC restart logic, ensuring Stop+Start cycles do not cause panics, data races, or goroutine leaks.

Defect fixes

Defect 1: Start() reuses the grpcServer created in the constructor. On the second Start() call, serve() triggers the gRPC fatal error RegisterService after Serve (calls os.Exit(1)).

Defect 2: Stop() does not wait for background goroutines to exit — it calls close(stopCh) and returns immediately. When Start() creates new goroutines that read the new stopCh, stale goroutines may still hold references to the old stopCh, causing a data race.

Defect 3: Stop() does not clean up the Unix socket file, so net.Listen may fail on restart.

Fix summary

File	Changes
server.go	1. Rebuild ps.grpcServer = grpc.NewServer() in Start() (Defect 1); 2. Add wg sync.WaitGroup to track all goroutines (Defect 2); 3. Stop(): close(stopCh) → grpcServer.Stop() → wg.Wait() → os.Remove(socket) (Defect 2, 3); 4. All three goroutines call wg.Add(1)/Done(); 5. Add private test hook fields (dialFunc, registerKubeletFunc, prepareHostResourcesFunc) for test isolation;
register.go	Add wg.Add(1)/Done() to watchAndRegister; add registerKubeletFunc and dialFunc hook checks;

Which issue(s) this PR fixes:

Fixes Project-HAMi/HAMi#1911

Special notes for your reviewer:

• All five restart tests correctly detect the original three defects (verified by reverting the fix and confirming test failures).
• When run under go test -race, the five restart tests may still report flaky false positives due to a Go 1.26 SSA optimization that replaces function parameter reads with direct struct field reads in go statement argument evaluation. See race-detector-false-positive.md for a full root-cause analysis.

Does this PR introduce a user-facing change?:

No

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository.

[archlitchi]

/lgtm

[hami-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: archlitchi, peachest

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [archlitchi]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment