Feature: attrd,crmd,libraries,tools: track resource failures per-operation

Previously, resource failures were tracked per resource, using
node attribute names such as "fail-count-myrsc".

Now, they are tracked per resource+operation+interval combination,
using multiple attributes such as "fail-count-myrsc#monitor_30000".

This does not change any failure handling behavior, but is preparation
for future changes.

This bumps the CRM feature set to be able to decide when to look for the new
attributes. Since it is impossible to convert any existing per-resource
attributes into per-operation ones, old attributes will simply be ignored once
the DC has the new feature set. This means that in a rolling upgrade, once
the last node is upgraded, it acts like a one-time cluster-wide timeout of all
past failures.