NodeLatencyStats API should use metav1.MicroTime instead of metav1.Time for send / receive timestamps

[antoninbas]

The API currently uses metav1.Time to represent timestamps:

antrea/pkg/apis/stats/v1alpha1/types.go

Lines 177 to 180 in 1907856

	// The timestamp of the last sent packet.
	LastSendTime metav1.Time `json:"lastSendTime,omitempty" protobuf:"bytes,2,opt,name=lastSendTime"`
	// The timestamp of the last received packet.
	LastRecvTime metav1.Time `json:"lastRecvTime,omitempty" protobuf:"bytes,3,opt,name=lastRecvTime"`

That means that when the timestamps are marshalled, "second-level" precision is used, which reduces the usefulness of the timestamps. In a cluster with low latency, we will typically end up with the same values for LastSendTime and LastRecvTime. These timestamps are not very important in themselves, and the latency data is provided by LastMeasuredRTTNanoseconds. The main reason to have these timestamps is to detect that a peer has stopped responding to probes.

In a future version of the API, we should consider updating the type for these timestamps to metav1.MicroTime, in which case they would be marshalled with "microsecond-level" precision.