Support floating point values in SLO evaluation

Monitoring values returned by Prometheus are always casted to ints. This was not an issue for monitoring metrics such as the consumer lag of stream processing systems, but does not work when having metrics with smaller values.