[ML] Lower Scale to Zero to 4 hours

[prwhelan]

Set default value to 4 hours to lower the dollar cost for most users, who only use inference intermittently, at the risk of cold start times.

Set a max value to 72 hours to prevent an effective nonzero min allocation.

[shubhaat]

Set a max value to 72 hours to prevent an effective nonzero min allocation.

Wait the intent here was to allow customers to be able to set it to a value between 4 and 72 hours. This seems like an internal setting that Elastic can set?