Fix up metric name conversion

[hoelzro]

Some metrics have substrings from the original metric name omitted from the resulting Prometheus metric - for example, both ClusterStatus.red and ClusterStatus.yellow become aws_es_cluster_status, which results in duplicate metric registration and a panic. This PR includes tests for various CloudWatch metric names, plus a new algorithm for splitting them up into words and recombining them.

[frioux]

Are the names for this backwards compatible or will the names change?

[hoelzro]

@frioux I initially thought not, but then I tried again and tested this against the names we use at work - the three that changed are:

• CPUUtilization was utilization, is now cpu_utilization
• LoginsLast10Minutes was logins_last10_minutes, is now logins_last_10_minutes
• LoginsLast10Minutes_all_count was logins_last10_minutes_all_count, is now logins_last_10_minutes_all_count

I'd argue that the first is a bug akin to the ClusterStatus.yellow one I set out to fix. Regarding the latter two, I didn't even see them when I listed metrics in CloudWatch (which is how I missed them in my initial testing), but I'm also fairly indifferent on last10 vs last_10.