AWS default machine type not always supported. (when VPC is targeted)

It seems for some credentials, or possibly just in some regions, m3.medium is no longer available.
I believe for 'juju bootstrap' we actually switched to t2.medium because of cpu vs cost issues.
I believe a lot of AWS has transitioned to M4.
But ultimately, we seem to be hard coding an instance type that used to be very common, but isn't guaranteed to be available, which means we don't fall back to a different type based on CPU/Mem/etc.
We should rethink how we do instance type selection, and possibly roll out changes to several different Juju versions.

Interestingly, it seems when using EC2 Classic m3.mediums *are* available.

The informative message from 2.2 should be back, as a result of fixing https://bugs.launchpad.net/juju/+bug/1732564.

We do not hard code instance types. We hard code minimum CPU/memory.

The problem is that the instance type is available in the region, but only in some availability zones, and apparently we're not retrying in all the zones for this error scenario.

This has been fixed since 2.3-rc1.

5 pending pending xenial failed to start machine 5 in zone "us-east-1c", retrying in 10s with new availability zone: cannot run instances: Your requested instance type (m3.medium) is not supported in your requested Availability Zone (us-east-1c). Please retry your request by not specifying an Availability Zone or choosing us-east-1a, us-east-1b, us-east-1e, us-east-1d. (Unsupported)

(and then it tried us-east-1f, and failed again with the same error; and then tried us-east-1b, and succeeded)