to the meaning of the individual metrics: the documentation only explains on a high-level and admittedly assume a detailed understanding of the inner workings of Elasticsearch. Btw, there is also http://esrally.readthedocs.io/en/stable/metrics.html that explains the raw metrics that Rally uses to create the summary report.
IMHO the primary metrics to look at are the request-related metrics: throughput (e.g. for bulk indexing operations) and service time or latency (for latency-sensitive operations like queries). The other metrics serve as secondary metrics to better understand where Elasticsearch spent time during indexing: was it busy merging a lot of segments, where merges throttled for a long time (disk issues?) etc. I think these can just serve as a first indication and once you see high values you need to dig further anyway with other tools like
vmstat, a Java profiler, ... .