Ergonomic settings for indices recovery based on node size for data_cold role

[dliappis]

To enhance the performance of the cold tier during failure scenarios or prewarming we could move away from the fixed 40MB/s value for indices.recovery/max_bytes_per_sec and instead ergonomically choose the value based on the node size.

So for nodes with the data_cold node_role we could implement a scheme like:

NOTE: the following table is referring to the available memory for the node (e.g. memory available inside a container) not jvm heap size

node size in GB	indices.recovery.max_bytes_per_sec	change from current default
node_size <= 1	40 MB/s	unchanged
1 < node_size <= 2	40 MB/s	unchanged
2 < node_size <= 4	40 MB/s	unchanged
4 < node_size <= 8	60 MB/s	increased
8 < node_size <= 16	90 MB/s	increased
16 < node_size <= 32	125 MB/s	increased
32 < node_size	250 MB/s	increased

[elasticmachine]

Pinging @elastic/es-distributed (Team:Distributed)