In TransportNodeDeprecationCheckAction we check for nodes exceeding the low watermark by passing in the (filtered) static settings from the local node, and the (filtered) dynamic settings from its cluster state, and reporting a violation of the watermark if the node's disk usage is too high according to either of these settings sets:

There's several issues with this:

1. There should be no facility for filtering out the relevant settings here.
2. The node-local settings have no relevance to disk watermarks unless the node is the elected master.
3. Even the elected master's node-local settings have no relevance to disk watermarks if the settings are overridden with dynamic values.
4. Conversely, if the settings are not overriden with dynamic values then checking filteredClusterState.metadata().settings() means checking against the default value of 85%, even if the master's node-local settings specify a different value.

Instead, since this action is originally invoked by the elected master, it should include in its request the actual disk watermark that the node should use.