[ML] Improve error message notifications when insufficient available memory

[sophiec20]

If an end-user has a cluster with 3x master, 3x data and 1x ML node (say), then if they cannot open an ML job due to running out of memory, then they get a pretty awful error message.

Each nodes returns to say that it cannot open the job because it is not an ML node, and the ML node returns to say it's full. We also get the additional message to say the datafeed could not start. See below.

Can we trap this and display a top level error? e.g. Insufficient available memory. More nodes or more memory required or close unused jobs.

I would expect the full error to be useful in troubleshooting and it could be hidden with an "expand to view full details".

Relates elastic/elasticsearch#29950, and possibly #17961

cc @skearns64

[elasticmachine]

Pinging @elastic/ml-ui

[droberts195]

elastic/elasticsearch#31925 moves the detailed message to a nested exception which will not be the root_cause of the open job failure.

This will work nicely if it is possible for Kibana to display the root_cause by default but also allow access to the nested exception message if desired. But if Kibana does not allow access to the nested exceptions then maybe it's not useful.

[jgowdyelastic]

#20880 raised to fix this.

Duplicate errors are grouped with the ability to expand to see the full response