Properly initialize metrics in rpcchainVM

[yacovm]

Why this should be merged

The RPC chain VM gRPC dispatcher initializes the metrics at Initialize() but it doesn't need any input from the Initialize method's parameters. It's better to initialize the metrics at the construction of the object.

This is needed because currently, if the gRPC service is requested to dispatch a gathering of the metrics, it might be that the metrics in the backend is yet to be initialized, and this is a problem.

How this works

Just a refactoring.

How this was tested

CI

[aaronbuchwald]

Left one nit, then this LGTM

[aaronbuchwald]

Can we switch to just inlining this for the smallest diff? ie. add metrics: metrics.NewPrefixGatherer() as the next line in the struct definition?

[darioush]

nit: could just be metrics: metrics.NewPrefixGatherer()

[aaronbuchwald]

This fixes a race condition while initializing an instance of the RPCChainVM that triggered a panic in AvalancheGo prior to this PR:

/go/pkg/mod/github.com/ava-labs/[email protected]/vms/rpcchainvm/vm_server.go:563 +0x1a

The path that triggered this bug is registering VMClient (reference held by the chain manager) as a gatherer here during vm.Initialize(...): https://github.com/ava-labs/avalanchego/blob/v1.11.11/vms/rpcchainvm/vm_client.go#L159

This is called before calling Initialize on the VM server with all of the required arguments from the vm client: https://github.com/ava-labs/avalanchego/blob/v1.11.11/vms/rpcchainvm/vm_client.go#L211. This is before the VM server would have initialized its own metrics field: https://github.com/ava-labs/avalanchego/blob/v1.11.11/vms/rpcchainvm/vm_server.go#L137.

This can trigger a panic because Gather can be called at any time, which can call Gather on the VMClient as soon as it registers itself and in the middle of vm.Initialize(...).