Active watches on kube apiserver in kubernetes filter plugin

**Is your feature request related to a problem? Please describe.**
In_tail plugin has latency checking logs from containers (files discovery). When pod/container fails to start, the moment event reaches filter plugin there can be no longer pod metadata in apiserver.

**Describe the solution you'd like**
Plugin can open watch on pods in apiserver and store them in local cache. There won't be situation when fluent-bit (plugin) won't know about pod.

**Describe alternatives you've considered**

**Additional context**
This was already introduced in https://github.com/fabric8io/fluent-plugin-kubernetes_metadata_filter and is enabled by default (`watch` property)

Start by locating the Kubernetes filter plugin and its pod-metadata lookup path; the issue does not name files or tests. Review the upstream fluent-plugin-kubernetes_metadata_filter watch behavior referenced in the issue, then determine how an active pod watch and local cache should cover metadata missing when a container-start event arrives. Done means the plugin retains and uses pod metadata in that case.

Written by the indexing model from the issue text.